CN120428215B - A multi-target automatic tracking method and system based on unmanned intelligent turntable - Google Patents
A multi-target automatic tracking method and system based on unmanned intelligent turntableInfo
- Publication number
- CN120428215B CN120428215B CN202510926166.4A CN202510926166A CN120428215B CN 120428215 B CN120428215 B CN 120428215B CN 202510926166 A CN202510926166 A CN 202510926166A CN 120428215 B CN120428215 B CN 120428215B
- Authority
- CN
- China
- Prior art keywords
- target
- track
- scanning
- data
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The application provides a multi-target automatic tracking method and system based on an unmanned intelligent turntable. The method comprises the steps of synchronously collecting electromagnetic echoes and environmental depth data through a detector to construct an observation data set, separating thermal radiation, surface textures and motion characteristics, establishing a four-dimensional characteristic space containing space-time coordinates, identifying a shielding area, dynamically adjusting signal parameters to enhance shielding area characteristics, feeding back optimized records, calculating an optimal scanning path based on parameter optimized values, driving a sensor array to scan at multiple angles to generate a weighted data set, combining historical track data, calculating a parallel optimized track prediction model through distributed computation, coordinating turntable parameter synchronization and decision, and finally realizing efficient reconstruction of multi-target continuous tracks in a shielding environment. According to the technical scheme provided by the application, the reconstruction of the multi-target track in the shielding environment is realized through multi-dimensional feature fusion and dynamic parameter optimization, so that the environmental adaptability and tracking efficiency of the unmanned system are improved.
Description
Technical Field
The application relates to the technical field of multi-target intelligent tracking in a shielding environment, in particular to a multi-target automatic tracking method and system based on an unmanned intelligent turntable.
Background
In civil scenes such as intelligent monitoring of urban transportation hubs, personnel management and control of large-scale venues, wild animal observation in natural protection areas and the like, an unmanned intelligent turntable system is required to realize uninterrupted tracking on multiple types of dynamic targets in a dense shielding environment. The core technical requirements of the application are that the system needs to have multi-mode data synchronization capability, electromagnetic sensing and environmental depth data are effectively fused to construct a unified space-time reference, the system needs to have self-adaptive feature extraction capability, identifiable features are dynamically strengthened when a target is partially shielded, the system needs to have collaborative sensing and track continuity guaranteeing capability, shielding period track prediction and reconstruction are realized through multi-angle scanning optimization and distributed computation, and the requirement of high-robustness tracking under a complex scene is met.
The current mainstream scheme for multi-target tracking in a dense shielding environment is a scheme based on single-sensor time sequence scanning and static characteristic matching. The scheme adopts a single high-precision sensor to scan a monitoring area according to a preset fixed path period. After the system recognizes the target through the background difference or motion detection algorithm, the system extracts a universal template pre-stored in the feature library to perform matching positioning. When the occlusion is detected, the system predicts the current position by adopting a linear motion model according to the historical track of the target, and keeps the original scanning path and parameters unchanged. All data processing and track calculation are completed in a centralized way by a processor, and scanning resources are distributed according to a preset priority rule.
The defects of the existing scheme are shown in a concentrated way, although a single sensor time sequence scanning mechanism can maintain basic monitoring coverage, multi-mode sensing data cannot be effectively fused in a dynamic shielding scene, meanwhile, a static feature library matching strategy can process conventional target recognition, dynamic feature enhancement capability is lacking when a target part is shielded, track fracture and information deletion are easy to occur only depending on an extrapolation position of a linear motion model, and more prominently, a centralized processor processing architecture simplifies data flow management, but faces to calculation requirements in multi-target concurrent tracking, calculation delay is larger, and target association error rate in a shielding period is increased.
Disclosure of Invention
The application provides a multi-target automatic tracking method and system based on an unmanned intelligent turntable, which are used for solving the problems of track fracture, characteristic loss and parameter synchronization lag in the prior art when multi-target tracking is performed in a dynamic shielding scene.
In a first aspect, the present application provides a multi-target automatic tracking method based on an unmanned intelligent turntable, including:
Acquiring electromagnetic echo signals and environment depth data of a target area through a detector of an unmanned intelligent turntable, and synchronously calibrating the electromagnetic echo signals and the environment depth data in a time dimension to form an observation data set;
Based on the physical characteristic difference of different sensors in the observation data set, separating out the heat radiation characteristics, the surface texture characteristics and the motion state characteristics of the target, constructing a four-dimensional characteristic space containing space-time coordinates, and simultaneously identifying the abnormal area of the surface of the target as a shielding area;
According to the target distribution density in the four-dimensional feature space and the feature conflict intensity of the shielding area, adjusting signal parameters of the detector to enhance the target feature of the shielding area, and feeding the adjusted signal parameters back to the four-dimensional feature space to form a feature enhancement record;
Calculating an optimal scanning path according to the signal parameter optimization value in the characteristic enhancement record and the acquired electromagnetic echo signal, and driving a sensor array to perform multi-angle scanning according to the optimal scanning path so as to generate a scanning data set with weight;
And establishing a target track prediction model based on the scanning data set and the historical track data, and performing parallel optimization on the target track prediction model by using distributed computing nodes so as to coordinate parameter synchronization and decision optimization of the unmanned intelligent turntable, thereby realizing reconstruction of multi-target continuous tracks in a shielding environment.
Optionally, the establishing a target track prediction model based on the scan data set and the historical track data, and performing parallel optimization on the target track prediction model by using a distributed computing node to coordinate parameter synchronization and decision optimization of the unmanned intelligent turntable, so as to realize reconstruction of multi-target continuous tracks in a shielding environment, including:
extracting position coordinates, time marks and path weight values from the scanning data set, forming a space-time sequence of the scanning data based on the position coordinates and the time marks, and converting the path weight values into track confidence weights;
aligning a position sequence in the historical track data with a space-time sequence of the scanning data, and establishing a dynamic displacement equation based on the track confidence weight to complete the construction of a target track prediction model;
After the target track prediction model is deployed to the distributed computing nodes, each node processes a target data subset and optimizes the target track prediction model by exchanging parameters of the dynamic displacement equation, so that the spatial deviation between a prediction position and scanning data is minimized;
Coordinating an unmanned intelligent turntable based on an optimization result to feed back the predicted position to a detector unit to adjust a scanning angle focus predicted area, and simultaneously comparing the track confidence weight with a set threshold value to adjust the signal transmitting power of the detector;
and iteratively executing the prediction process of the target track prediction model and the parameter updating process of the dynamic displacement equation under the shielding environment, and generating a continuous track sequence by calculating a connection track break point through extrapolation of the dynamic displacement equation.
Optionally, the generating a continuous track sequence by calculating a connection track break point through extrapolation of the dynamic displacement equation includes:
when the track interruption is detected, recording the last effective position coordinate as an initial interruption point and recording the first recapture position coordinate as a termination interruption point;
The initial break point is taken as an initial position, the dynamic displacement equation is applied to predict the position of the next time point, the predicted position is taken as a virtual scanning point to be inserted into a scanning data set, attenuation type track confidence weights are given to form an expansion data set, and parameters of the dynamic displacement equation are updated based on the expansion data set;
forward iterating and predicting the initial break point to an intermediate time point according to the time sequence, taking the position of the intermediate time point at the moment as a forward predicted position, and reversely iterating and predicting the same intermediate time point according to the time sequence from the termination break point at the same time, and taking the position of the intermediate time point at the moment as a reverse predicted position;
And when the spatial deviation between the forward predicted position and the reverse predicted position is smaller than a tolerance threshold, connecting the initial break point, the intermediate connection point and the termination break point to form a continuous track segment.
Optionally, the calculating an optimal scanning path according to the signal parameter optimization value in the feature enhancement record and in combination with the acquired electromagnetic echo signal, and driving the sensor array to perform multi-angle scanning according to the optimal scanning path, so as to generate a weighted scanning data set, includes:
generating a region scanning priority based on the characteristic enhancement value in the characteristic enhancement record, wherein a region with higher characteristic enhancement value corresponds to a higher scanning priority;
Identifying a low signal intensity region and a high priority region with the scanning priority larger than a preset threshold by combining the intensity distribution of the electromagnetic echo signals, taking the highest priority region as a path starting point, forcibly covering all the high priority region and the low signal intensity region, simultaneously realizing the minimization of the total steering angle, and generating an optimal scanning path;
Generating a steering angle sequence of the detector unit based on the optimal scanning path, and driving the sensor array to scan according to the steering angle sequence to obtain the position coordinate, the time mark, the electromagnetic echo signal value and the environment depth value of each scanning point;
And distributing a path weight value equal to the region characteristic enhancement value for each scanning point, integrating the information of all the scanning points, and forming a scanning data set containing the position coordinates, the time marks, the electromagnetic echo signal values, the environment depth values and the path weight values.
Optionally, the separating the thermal radiation feature, the surface texture feature and the motion state feature of the target based on the physical characteristic difference of different sensors in the observation dataset, and constructing a four-dimensional feature space including space-time coordinates, and identifying the abnormal area of the surface of the target as the shielding area, includes:
Extracting electromagnetic echo signal values and environment depth values and corresponding time marks and position coordinates based on physical characteristic differences of different sensors in the observed dataset;
based on the change mode of the electromagnetic echo signal value, the thermal radiation characteristic is separated through signal intensity threshold comparison, the surface texture characteristic is separated through adjacent depth value difference calculation based on the local fluctuation mode of the environment depth value, and the motion state characteristic is separated through position deviation calculation of continuous time points based on the change trend of the position coordinates and the time marks;
Respectively associating corresponding time marks and position coordinates according to the heat radiation characteristics, the surface texture characteristics and the motion state characteristics to construct a four-dimensional characteristic space;
In the four-dimensional feature space, an abnormal region of the target surface is identified through abrupt change detection of the electromagnetic echo signal value and the environment depth value, and when the surface texture feature or the heat radiation feature of the abnormal region has abnormal decline of the intensity value in a local region, the abnormal region is marked as a shielding region.
Optionally, the adjusting the signal parameter of the detector according to the target distribution density in the four-dimensional feature space and the feature collision strength of the shielding area to enhance the target feature of the shielding area, and feeding back the adjusted signal parameter to the four-dimensional feature space to form a feature enhancement record, includes:
counting the number of the feature points in the unit space area according to the aggregation degree of the position coordinates in the four-dimensional feature space so as to calculate the target distribution density;
in the shielding region, respectively comparing intensity values of the surface texture features and the heat radiation features in a normal region and the shielding region based on intensity value deviation of the surface texture features and the heat radiation features so as to calculate corresponding feature collision intensity;
according to the target distribution density and the characteristic collision intensity, adjusting signal parameters of the detector through a signal parameter optimization rule, and enhancing target characteristics of the shielding area;
And applying the adjusted signal parameters to a detector, re-acquiring signals of the shielding area, updating the numerical values of the target distribution density and the characteristic collision intensity in the four-dimensional characteristic space, and recording the adjustment process of the signal parameters, the target distribution density and the characteristic collision intensity to form a characteristic enhancement record.
Optionally, the acquiring, by the detector of the unmanned intelligent turntable, the electromagnetic echo signal and the environmental depth data of the target area, and calibrating the electromagnetic echo signal and the environmental depth data in synchronization in a time dimension, to form an observation data set includes:
Setting a detector group comprising an electromagnetic sensor and a depth sensor on the unmanned intelligent turntable, wherein the electromagnetic sensor emits electromagnetic waves at fixed time intervals and receives reflected signals to generate electromagnetic echo signals, and the depth sensor acquires depth information of a target area as environmental depth data based on an optical principle;
Automatically attaching a time stamp uniformly managed by an internal clock of the unmanned intelligent turntable when the electromagnetic sensor and the depth sensor generate data, and decomposing the electromagnetic echo signal and the environmental depth data into discrete data points containing the time stamp, the position coordinates and corresponding signal intensity values;
and performing one-to-one pairing on the electromagnetic echo signals and discrete data points of the environmental depth data based on the time markers to form data units, and integrating the data units into a structured data set comprising the time markers, the position coordinates and corresponding signal intensity values as an observation data set.
In a second aspect, the present application provides a multi-target automatic tracking system based on an unmanned intelligent turntable, comprising:
The acquisition module is used for acquiring electromagnetic echo signals and environment depth data of a target area through a detector of the unmanned intelligent turntable, and synchronously calibrating the electromagnetic echo signals and the environment depth data in a time dimension to form an observation data set;
The construction module is used for separating out the thermal radiation characteristics, the surface texture characteristics and the motion state characteristics of the target based on the physical characteristic differences of different sensors in the observation data set, constructing a four-dimensional characteristic space containing space-time coordinates, and identifying the abnormal area of the surface of the target as a shielding area;
the adjusting module is used for adjusting signal parameters of the detector to strengthen the target characteristics of the shielding area according to the target distribution density in the four-dimensional characteristic space and the characteristic conflict intensity of the shielding area, and feeding the adjusted signal parameters back to the four-dimensional characteristic space to form a characteristic enhancement record;
the scanning module is used for driving the sensor array to perform multi-angle scanning, calculating an optimal scanning path according to the signal parameter optimization value in the characteristic enhancement record and the acquired electromagnetic echo signal so as to generate a scanning data set with weight;
and the prediction module is used for establishing a target track prediction model based on the scanning data set and the historical track data, and carrying out parallel optimization on the target track prediction model by utilizing a distributed computing node so as to coordinate the parameter synchronization and decision optimization of the unmanned intelligent turntable and realize the reconstruction of the multi-target continuous track in the shielding environment.
In a third aspect, the application provides a computing device, which comprises a processing component and a storage component, wherein the storage component stores one or more computer instructions, and the one or more computer instructions are used for being called and executed by the processing component to realize the multi-target automatic tracking method based on the unmanned intelligent turntable.
In a fourth aspect, the present application provides a computer storage medium storing a computer program, where the computer program is executed by a computer to implement a multi-objective automatic tracking method based on an unmanned intelligent turret according to the first aspect.
In the embodiment of the application, an electromagnetic echo signal and environmental depth data of a target area are acquired through a detector of an unmanned intelligent turntable, the electromagnetic echo signal and the environmental depth data are synchronously calibrated in a time dimension to form an observation data set, the thermal radiation characteristic, the surface texture characteristic and the motion state characteristic of a target are separated based on physical characteristic differences of different sensors in the observation data set, a four-dimensional characteristic space containing space-time coordinates is constructed, an abnormal area of the surface of the target is identified as an occlusion area, signal parameters of the detector are adjusted according to the target distribution density in the four-dimensional characteristic space and the characteristic collision intensity of the occlusion area, the target characteristics of the occlusion area are enhanced, the adjusted signal parameters are fed back to the four-dimensional characteristic space to form a characteristic enhancement record, an optimal scanning path is calculated according to signal parameter optimization values in the characteristic enhancement record, the sensor array is driven according to the optimal scanning path to generate a multi-angle scanning data set, a target track prediction model is established based on the scanning data set and historical track data, and the target track prediction model is used for carrying out multi-angle synchronous reconstruction on the target-level prediction model, and the optimal environment is realized under the condition of the optimal environment model.
The technical scheme of the application has the following beneficial effects:
According to the method, through time synchronous calibration of multi-mode sensing data and dynamic construction of four-dimensional feature space, the technical problems of target feature degradation and track fracture in a complex shielding environment are effectively solved. The multi-angle scanning strategy based on signal parameter optimization driving and the distributed parallel optimization model realize the cooperative processing of dynamic adjustment of a scanning path and track prediction, overcome the performance bottleneck of the traditional single-sensor scanning frame in a dense shielding scene, and ensure the integrity and stability of multi-target continuous track reconstruction.
The method comprises the steps of obtaining a path weight value of a scanning data set, establishing a path weight value, establishing a path confidence weight value, establishing a path time sequence, establishing a dynamic displacement equation based on the path confidence weight, establishing a target path prediction model, deploying the model to distributed computing nodes, processing a data subset in parallel by each node, exchanging parameters to optimize the model, minimizing the spatial deviation between a prediction position and the scanning, dynamically adjusting the scanning angle of a turntable and the signal power of a detector according to the optimization result, simultaneously iteratively executing prediction and parameter updating in a shielding environment, and extrapolating a connection path break point through the dynamic displacement equation to generate a continuous path sequence. The method solves the technical problems of multi-target track fracture and association errors in the shielding environment by distributed parallel optimization and iterative updating of a dynamic displacement equation, improves track prediction performance and continuity, improves track reconstruction integrity, and effectively reduces target omission ratio in the dense shielding environment by a signal power dynamic adjustment mechanism based on scanning confidence weight, thereby guaranteeing multi-target tracking robustness in the complex environment.
These and other aspects of the application will be more readily apparent from the following description of the embodiments.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 shows a flow chart of a multi-target automatic tracking method based on an unmanned intelligent turntable;
FIG. 2 shows a scene diagram of a multi-target automatic tracking method based on an unmanned intelligent turntable;
fig. 3 shows a schematic structural diagram of a multi-target automatic tracking system based on an unmanned intelligent turntable;
FIG. 4 illustrates a schematic diagram of a computing device provided by the present application.
Detailed Description
In order to enable those skilled in the art to better understand the present application, the following description will make clear and complete descriptions of the technical solutions according to the embodiments of the present application with reference to the accompanying drawings.
In some of the flows described in the specification and claims of the present application and in the foregoing figures, a plurality of operations occurring in a particular order are included, but it should be understood that the operations may be performed out of order or performed in parallel, with the order of operations such as 101, 102, etc., being merely used to distinguish between the various operations, the order of the operations themselves not representing any order of execution. In addition, the flows may include more or fewer operations, and the operations may be performed sequentially or in parallel. It should be noted that, the descriptions of "first" and "second" herein are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, and are not limited to the "first" and the "second" being different types.
Researches show that the existing intensive shielding environment multi-target tracking scheme relies on a single-sensor time sequence scanning and static feature matching mechanism, a fixed path scanning mode is difficult to fuse multi-mode time-space information of electromagnetic echoes and environment depth data, so that feature dislocation under complex shielding scenes such as building group gaps is caused, a static feature library matching strategy cannot dynamically strengthen local identifiable features of shielding targets, an extrapolation track of a linear motion model is easy to break, a centralized processor architecture is insufficient in calculation force during multi-target concurrent tracking, errors of association between scanning parameter optimization delay and shielding period targets are easy to cause, and reliability of track continuity reconstruction is restricted.
Aiming at the problems, the invention provides a multi-target automatic tracking method based on an unmanned intelligent turntable, which is characterized in that a dynamically optimizable four-dimensional feature space is constructed through time-space synchronous calibration of electromagnetic echo and environmental depth data, and scanning parameter self-adaptive adjustment and track prediction model parallel optimization are realized based on distributed computation. According to the method, the problem of single-sensor information limitation is solved through multi-mode sensing data space-time fusion, the characteristic degradation defect of static template matching in a shielding scene is overcome by a dynamic characteristic enhancement mechanism, the track prediction delay is reduced to the millisecond level by a distributed parallel optimization architecture, the calculation efficiency is improved compared with a traditional centralized processing architecture, the accuracy of target characteristic extraction of a shielding area is improved through signal parameter self-adaptive adjustment, and the reliability and the integrity of continuous track reconstruction in a complex scene are further improved.
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the application.
Fig. 1 is a flowchart of a multi-target automatic tracking method based on an unmanned intelligent turntable according to an embodiment of the present application, as shown in fig. 1, the method includes:
101. Acquiring electromagnetic echo signals and environment depth data of a target area through a detector of an unmanned intelligent turntable, and synchronously calibrating the electromagnetic echo signals and the environment depth data in a time dimension to form an observation data set;
Optionally, step 101 may specifically include the following steps:
1011. Setting a detector group comprising an electromagnetic sensor and a depth sensor on the unmanned intelligent turntable, wherein the electromagnetic sensor emits electromagnetic waves at fixed time intervals and receives reflected signals to generate electromagnetic echo signals, and the depth sensor acquires depth information of a target area as environmental depth data based on an optical principle;
1012. Automatically attaching a time stamp uniformly managed by an internal clock of the unmanned intelligent turntable when the electromagnetic sensor and the depth sensor generate data, and decomposing the electromagnetic echo signal and the environmental depth data into discrete data points containing the time stamp, the position coordinates and corresponding signal intensity values;
1013. And performing one-to-one pairing on the electromagnetic echo signals and discrete data points of the environmental depth data based on the time markers to form data units, and integrating the data units into a structured data set comprising the time markers, the position coordinates and corresponding signal intensity values as an observation data set.
In the above scheme, the electromagnetic echo signal refers to a time-varying signal generated by the electromagnetic sensor transmitting electromagnetic waves and receiving a target reflected signal, and includes position coordinates, motion state and material characteristic information of the target, which can be used for target detection, classification and tracking. The environment depth data is discrete point cloud data obtained based on an optical principle, reflects the space distance between a target and a sensor and a scene three-dimensional structure, and can be used for constructing unified space-time reference and obstacle modeling. Discrete data points refer to atomic data units consisting of time stamps, spatial coordinates, and signal strength parameters, providing structured input for feature extraction and multi-sensor correlation analysis. The data unit is used for reflecting the physical state and the spatial distribution characteristic of the target at a specific moment through the combination of the electromagnetic echo signals and the environment depth data points associated with the time stamps and supporting cross-modal data fusion. The structured dataset refers to a multi-dimensional observation dataset stored in a unified format, comprising electromagnetic features, geometric features and environmental context information aligned in time series, supporting efficient querying and distributed computing.
In the embodiment of the present application, firstly, through step 1011, a detector group composed of an electromagnetic sensor and a depth sensor is installed on a mechanical arm or a movable base of an unmanned intelligent turntable. Electromagnetic sensors generate electromagnetic echo signals by periodically transmitting electromagnetic pulses and receiving reflected signals from a target area, and their operating frequency and transmit power can be dynamically adjusted according to environmental complexity, for example, metal detection scenes employ high-frequency narrow beams. The depth sensor acquires a depth profile of the target region based on optical principles, generating environmental depth data reflecting spatial geometrical features, such as structured light projection or ToF time of flight measurements. The electromagnetic sensor and the depth sensor are physically connected through a hardware interface of the turntable control module, and the time sequence consistency of electromagnetic emission and depth sampling is ensured through the synchronous signal trigger, for example, after each pulse is emitted by the electromagnetic sensor, the depth sensor immediately starts single-frame acquisition, so that data dislocation caused by time difference is avoided.
Then, when the electromagnetic sensor and the depth sensor generate raw data, a time stamp generated by a high-precision clock with the precision reaching the microsecond level inside the turntable is automatically added to form a raw data stream with a time mark through step 1012. For example, the electromagnetic echo signal is recorded as "timestamp+received intensity+azimuth/elevation angle", and the environmental depth data is marked as "timestamp+pixel coordinates+depth value". The system then splits the continuous signal stream into discrete data points by a data parsing module at time stamps, each data point containing time information, spatial coordinates, and gray values of amplitude or depth values of electromagnetic echoes of the corresponding signals. The process is realized through a preset analysis protocol, for example, a JSON format is adopted to package data fields, so that the compatibility of data structures of different sensors is ensured.
Finally, the electromagnetic echo data is matched with the environmental depth data based on the uniform time stamp, for example, the electromagnetic echo intensity value of 50dB with the time stamp t=0.5 s and the depth data (coordinate x=1.2 m, y=3.4 m, depth 2.1 m) at the same time form one data unit, via step 1013. In the matching process, the system processes short dislocation caused by sensor delay through a sliding window algorithm or a time tolerance threshold value, and eliminates non-matching items. Finally, all data units are integrated into a structured data set in time sequence as an observation data set, and standardized fields are adoptedThe storage supports direct input machine learning models or three-dimensional reconstruction algorithms. For example, an observation dataset generated by a scan may contain thousands of records, each of which accurately describes the location, electromagnetic properties, and spatial depth of a target at a particular time, providing complete input for subsequent environmental modeling and target identification.
In practical application, in an industrial warehouse inspection scene, an unmanned intelligent turntable is carried with a detector group consisting of a 24GHz millimeter wave radar and a structured light depth camera, and data of a target area is synchronously acquired at a rate of 20 frames per second. When the turntable scans the shelf area, the millimeter wave radar emits cone beams at intervals of 50ms to capture electromagnetic echo signals of the metal shell of the mobile AGV, and the depth camera generates 1280 multiplied by 720 depth point clouds through infrared structured light projection. The system ensures that radar pulse emission is aligned with depth camera exposure strictly through a microsecond synchronous trigger, and all data are added with accurate time stamps provided by a turntable main control chip. The data analysis module splits the radar echo into discrete points including time stamps, polar coordinates and signal strength, and simultaneously analyzes the depth data into time stamps, pixel coordinates and depth values. The radar data point intensity 61dB at the time of t= 1622543950.124000 is dynamically associated with the same-frame depth point depth of 3.02m, the world coordinate x=2.31 m, y=1.87 m and z=3.02 m through a sliding window matching algorithm, so as to form a structured data unit. A scan lasting 10 seconds generates a total of 200 frames of matching data sets, each record containing standardized fields: . The data set successfully fuses electromagnetic reflection characteristics and three-dimensional space positions of the AGV, provides cross-mode input for subsequent multi-target motion trail reconstruction, and has an actual measurement positioning error smaller than 0.1m.
According to the whole scheme of 101, depth-related sensing of electromagnetic characteristics and geometric forms is realized through the technology of collaborative acquisition and high-precision space-time synchronization of the multi-mode sensor. The electromagnetic sensor and the depth sensor form a detector group on the unmanned turntable, microsecond time alignment is realized based on a unified clock source, the time dislocation problem of cross-mode data is eliminated, the continuous signal flow is disassembled into discrete data points with time stamps by the data analysis engine, the short delay among the sensors is processed by a dynamic time warping algorithm, the equipment installation difference is compensated by a spatial interpolation algorithm, and finally the matched multidimensional data units are integrated into a standardized observation data set according to time sequence. According to the scheme, through space-time consistency calibration and structural fusion, the target identification accuracy and the system decision reliability in a complex industrial scene are remarkably improved, and high-fidelity input is provided for dynamic obstacle track prediction and three-dimensional environment modeling.
102. Based on the physical characteristic difference of different sensors in the observation data set, separating out the heat radiation characteristics, the surface texture characteristics and the motion state characteristics of the target, constructing a four-dimensional characteristic space containing space-time coordinates, and simultaneously identifying the abnormal area of the surface of the target as a shielding area;
Optionally, step 102 may specifically include the following steps:
1021. extracting electromagnetic echo signal values and environment depth values and corresponding time marks and position coordinates based on physical characteristic differences of different sensors in the observed dataset;
1022. Based on the change mode of the electromagnetic echo signal value, the thermal radiation characteristic is separated through signal intensity threshold comparison, the surface texture characteristic is separated through adjacent depth value difference calculation based on the local fluctuation mode of the environment depth value, and the motion state characteristic is separated through position deviation calculation of continuous time points based on the change trend of the position coordinates and the time marks;
1023. respectively associating corresponding time marks and position coordinates according to the heat radiation characteristics, the surface texture characteristics and the motion state characteristics to construct a four-dimensional characteristic space;
1024. in the four-dimensional feature space, an abnormal region of the target surface is identified through abrupt change detection of the electromagnetic echo signal value and the environment depth value, and when the surface texture feature or the heat radiation feature of the abnormal region has abnormal decline of the intensity value in a local region, the abnormal region is marked as a shielding region.
In the above scheme, the heat radiation characteristic refers to a dynamic signal reflecting the temperature distribution characteristic of the target surface, including temperature intensity and spatial distribution parameters, and can be used for identifying local overheating or low-temperature areas. The surface texture features refer to dynamic signals reflecting the geometric deformation characteristics of the target surface, and include local gradient amplitude and texture complexity parameters, and can be used for detecting surface scratches or concave-convex deformation. The motion state features are dynamic signals reflecting the spatial displacement characteristics of the target, and comprise motion vectors and track curvature parameters, and can be used for tracking the motion path and speed change of the target. The shielding region is an abnormal signal region reflecting the shielding characteristic of the target surface, comprises electromagnetic intensity dip and depth gradient abrupt change parameters, and can be used for marking obstacles or self-shielding parts in the field of view of the sensor.
In the embodiment of the present application, first, in step 1021, electromagnetic echo signal values, environmental depth values, corresponding time stamps and position coordinates are extracted from the observation data set by analyzing physical characteristic differences of the electromagnetic sensor and the environmental depth sensor, such as electromagnetic wave reflection characteristics and optical imaging principles. For example, from the observation dataset to formIs described herein).
Next, a differentiation algorithm is employed to isolate the target features based on the sensor physical property differences, via step 1022. The method comprises the steps of performing moving average filtering on electromagnetic echo signals to eliminate transient noise, separating high-intensity signals of a temperature abnormal region, such as 55dB heat radiation characteristics of the surface of a metal workpiece, based on a dynamic threshold value, performing Gaussian smoothing on an environment depth value to inhibit high-frequency noise, calculating local gradient amplitude by adopting a Sobel operator, reserving pixels with gradient values exceeding 5cm/m as surface texture mutation regions, such as depth gradient mutation caused by metal scratches, calculating continuous inter-frame displacement vectors by a Lucas-Kanade optical flow method, predicting a target motion track by combining Kalman filtering, and outputting a speed vectorAnd track curvature parameters, such as 0.1m/s transverse displacement and 0.05m/s longitudinal displacement generated by the rotary motion of the workpiece, and finally realizing the efficient separation of the heat radiation characteristic, the surface texture characteristic and the motion state characteristic.
Then, the separated features are related according to time-space coordinates in step 1023 to construct a space-time coordinate (x, y, z) electromagnetic intensityDepth gradient and motion vectorIs described. The storage format is defined by structuring the storage format definition fields t, x, y, z,,,,And unified characterization of the multi-source heterogeneous data is realized. For example, a certain frame data is stored as: And high-dimensional feature input is provided for target identification and behavior prediction in a follow-up complex scene.
Finally, in the four-dimensional feature space, the target surface occlusion region is identified by an anomaly detection algorithm, via step 1024. And performing first-order differential operation on the electromagnetic intensity and the environmental depth gradient sequence, judging the marked abrupt change points by a dynamic threshold value, then polymerizing adjacent abrupt change points by adopting morphological closing operation to form a continuous shielding region outline, and distinguishing self shielding from external shielding by a surface texture gradient or thermal radiation intensity verification mechanism. For example, when the edge of a metal workpiece is suddenly reduced from 60dB to 30dB in electromagnetic intensity and the depth gradient suddenly changes from 5cm/m to 20cm/m due to shielding by a sensor bracket, the system marks the edge as a shielding area through multi-dimensional feature verification, and the information can be used for dynamic obstacle avoidance by a path planning module.
In practical application, in an industrial quality inspection scene, a deployed unmanned turntable is provided with a high-frequency electromagnetic sensor and a ToF depth camera, and microsecond time alignment is realized through a synchronous controller. The electromagnetic sensor scans the surface of the metal casting at 20Hz frequency to generate a time-varying electromagnetic echo signal containing heat radiation characteristics, and the depth camera synchronously acquires three-dimensional point cloud data with the accuracy of 0.05 mm. The turret control module appends a 200ns precision time stamp to each frame of data, e.g. electromagnetic echo is marked asDepth data is marked as. The data parsing module splits the continuous signal into discrete points, matches the time-stamp aligned electromagnetic data points (1625000000.123 ns,62 dB) with the depth data points (1625000000.123 ns,3.2 mm) by a sliding window algorithm, forming a data unit containing space-time coordinates and multidimensional features. Separating heat radiation characteristics by adopting a dynamic threshold value, carrying out 5-point moving average filtering on electromagnetic signals, setting a 58dB threshold value to separate 62 dB+/-2 dB of a local overheating area on the surface of a casting, calculating a depth gradient by using a Sobel operator, identifying a surface texture mutation area of a casting air hole defect with the gradient of more than 8mm/m, calculating continuous frame-to-frame displacement vectors by a Lucas-Kanade optical flow method, and outputting velocity vectors (0.15 m/s,0.08 m/s) and curvature parameters by combining Kalman filtering prediction tracks. The four-dimensional feature space constructed includes time, space coordinates, electromagnetic intensity, depth gradient and motion vector, for example, a defect point is stored as. In a four-dimensional space, the electromagnetic intensity dip is detected through a first-order difference from 62dB to 32dB and the depth gradient mutation is detected from 8mm/m to 25mm/m, and a morphological closed operation polymerization abnormal region is combined and marked as a phi 15cm circular shielding region formed by shielding a sensor bracket.
According to the overall scheme of 102, through the multi-mode sensor data fusion and characteristic decoupling technology, based on the physical characteristic difference between electromagnetic echo and environmental depth values, efficient separation of heat radiation, surface texture and motion state characteristics is achieved, and a space-time aligned four-dimensional characteristic space is constructed. Through dynamic threshold detection and morphological analysis, the abnormal area of the target surface is accurately identified in the four-dimensional feature space, and the self-shielding and the external shielding are distinguished by combining a multi-dimensional feature intensity verification mechanism. The method effectively fuses multi-source heterogeneous data, breaks through the perception limitation of a single sensor in a complex industrial scene, remarkably improves the target recognition precision and obstacle avoidance response efficiency, and provides robust perception support for high-precision motion control.
103. According to the target distribution density in the four-dimensional feature space and the feature conflict intensity of the shielding area, adjusting signal parameters of the detector to enhance the target feature of the shielding area, and feeding the adjusted signal parameters back to the four-dimensional feature space to form a feature enhancement record;
optionally, step 103 may specifically include the following steps:
1031. Counting the number of the feature points in the unit space area according to the aggregation degree of the position coordinates in the four-dimensional feature space so as to calculate the target distribution density;
1032. in the shielding region, respectively comparing intensity values of the surface texture features and the heat radiation features in a normal region and the shielding region based on intensity value deviation of the surface texture features and the heat radiation features so as to calculate corresponding feature collision intensity;
1033. According to the target distribution density and the characteristic collision intensity, adjusting signal parameters of the detector through a signal parameter optimization rule, and enhancing target characteristics of the shielding area;
1034. And applying the adjusted signal parameters to a detector, re-acquiring signals of the shielding area, updating the numerical values of the target distribution density and the characteristic collision intensity in the four-dimensional characteristic space, and recording the adjustment process of the signal parameters, the target distribution density and the characteristic collision intensity to form a characteristic enhancement record.
In the above scheme, the target distribution density refers to the aggregation degree of target feature points in a unit volume in the four-dimensional feature space, and the spatial distribution characteristics of the target are quantified by counting the number distribution of the feature points such as electromagnetic intensity, depth gradient and the like in a specific area. The characteristic collision strength refers to the degree of difference between the shielding area and the normal area on the characteristics of surface texture, heat radiation and the like, and characteristic abnormality caused by shielding is quantified by comparing the mean value, variance and gradient change of the multidimensional characteristic values. The target features refer to a multi-dimensional physical attribute set of a target in a four-dimensional feature space, and the multi-dimensional physical attribute set comprises electromagnetic echo intensity, environmental depth gradient, motion state parameters and the like, and reflects the geometric structure and dynamic behavior of the target. The characteristic enhancement record refers to a closed loop feedback log formed by combining the change of target distribution density and characteristic conflict intensity in the re-acquired data through a parameter configuration sequence generated after the signal parameters of the detector are adjusted.
In the embodiment of the application, firstly, based on the aggregation degree of the position coordinates (x, y, z) in the four-dimensional feature space, the space is divided into units with fixed size (such as 0.1m×0.1 m) by a grid division method, the space resolution of each unit is ensured to be consistent by adopting an equidistant region division method by grid division, the number of electromagnetic intensity and depth gradient feature points in each grid unit is counted, the statistical result is smoothed by adopting a Gaussian kernel density estimation method, and the discrete noise is eliminated to generate target distribution density. Specifically, the contribution of each feature point is weighted by a kernel function, and a continuous density distribution map is generated. For example, in industrial detection, a defect area on the surface of a metal workpiece forms high-density clusters due to dense electromagnetic echoes, and a density thermodynamic diagram can visually display defect concentration areas. The process provides a spatial reference for subsequent parameter optimization by quantifying the spatial distribution characteristics.
Next, in the occlusion region, the surface texture gradient and the thermal radiation intensity are extracted, respectively, and the occlusion region is separated from the normal region by morphological operations, for example, marking the stent occlusion range based on connected region analysis, via step 1032. Comparing the shielding area with the characteristic value of the same target in a normal state, and calculating the gradient amplitude difference of the surface texture, wherein the formula is as follows: meanwhile, the difference and variance change of the heat radiation intensity are calculated, and the difference formula of the heat radiation intensity is as follows: . And adopting weighted Euclidean distance to fuse the multidimensional difference to generate a conflict intensity matrix. For example, sensor mount occlusion results in a texture gradient from 50 to 15, a thermal radiation intensity from 60dB to 28dB, and a collision intensity of 0.7. The process accurately distinguishes the self-shielding and the external shielding through multi-feature comparison.
The detector parameters are then adjusted using a dynamic weight distribution algorithm according to the target distribution density and the characteristic collision intensity, via step 1033. The high-density region reduces electromagnetic sensor gain to avoid signal overexposure, the low-density region increases gain to enhance weak signals, the high-conflict region relaxes the filtering threshold to preserve abnormal features, e.g., the defect dense region reduces gain to avoid signal overexposure, the background noise region increases gain to enhance weak signals, the high-conflict region relaxes the filtering threshold to adjust the mean filtering window from 5 x 5 to 3 x 3 preserving abnormal features. And predicting a target motion trail based on Kalman filtering, and adjusting the sampling frequency to match the target dynamic characteristics. And optimizing parameter combinations through a genetic algorithm, maximizing the characteristic identifiability of an objective function, and outputting the adjusted gain value and the sampling interval. The process dynamically adapts to environmental changes through a closed loop feedback mechanism.
Finally, the adjusted signal parameters are applied to the detector to re-acquire the electromagnetic echo and depth data of the shielding area, the new data and the original data are fused by utilizing an incremental learning algorithm, the density thermodynamic diagram in the four-dimensional feature space is dynamically updated, the change of the collision intensity is quantified through differential operation, for example, the adjusted electromagnetic intensity is restored to 55dB from 45dB, the collision intensity is reduced to 0.5, then the parameter adjustment sequence, the density trend diagram and the collision comparison table are stored in a log in a Protobuf format, an enhancement file containing a multi-dimensional feature change record is formed, the file supports the follow-up algorithm to realize self-adaptive parameter optimization through backtracking analysis, for example, the sensitivity attenuation rule of the sensor is rapidly positioned through the history record in industrial detection, and the target identification robustness under complex shielding scenes is improved.
In practical application, in an unmanned turntable bearing defect detection scene, a system discovers that a bearing raceway area has high target distribution density based on four-dimensional feature space analysis, characteristic points in a grid 0.05m < 3 >, the average value of texture gradients in a normal area of a bracket rotation shielding area is 45cm/m, heat radiation is 58dB, the shielding area suddenly drops to 12cm/m and 32dB, the characteristic collision intensity is 0.8, laser radar and millimeter wave radar parameters are dynamically adjusted according to the characteristic gradient, the laser sampling rate is increased from 10Hz to 30Hz to capture surface microcracks under the condition that a turntable rotates at a high speed of 2 pi rad/s, millimeter wave emission power is increased from 20dBm to 28dBm to penetrate greasy dirt shielding, the electromagnetic intensity of the shielding area is restored to 53dB after re-acquisition, the texture gradient is increased to 38cm/m, the collision intensity is reduced to 0.3, and an enhanced record is generated by updating the feature space。
According to the whole scheme of 103, intelligent closed-loop optimization of the signal parameters of the detector is realized by dynamically sensing the target distribution density in the four-dimensional feature space and accurately diagnosing the collision strength between the surface texture of the shielding region and the heat radiation features. The method can effectively identify the target gathering area, locate the characteristic distortion source caused by shielding, adaptively adjust signal parameters according to the characteristic distortion source, remarkably strengthen the characteristic expression of the target in the shielding area, and improve the detectability and the identification degree of the target in a complex environment. And immediately feeding the optimized parameters back to the detector for signal re-acquisition, updating the feature space data, verifying the enhancement effect, and simultaneously completely recording the complete processes of parameter adjustment, distribution density change and conflict strength evolution to form a traceable feature enhancement history. Finally, the mechanism continuously improves the target detection robustness of the system under the shielding scene, and provides reliable data support and decision basis for subsequent analysis optimization.
104. Calculating an optimal scanning path according to the signal parameter optimization value in the characteristic enhancement record and the acquired electromagnetic echo signal, and driving a sensor array to perform multi-angle scanning according to the optimal scanning path so as to generate a scanning data set with weight;
optionally, the step 104 may specifically include the following steps:
1041. Generating a region scanning priority based on the characteristic enhancement value in the characteristic enhancement record, wherein a region with higher characteristic enhancement value corresponds to a higher scanning priority;
1042. Identifying a low signal intensity region and a high priority region with the scanning priority larger than a preset threshold by combining the intensity distribution of the electromagnetic echo signals, taking the highest priority region as a path starting point, forcibly covering all the high priority region and the low signal intensity region, simultaneously realizing the minimization of the total steering angle, and generating an optimal scanning path;
1043. Generating a steering angle sequence of the detector unit based on the optimal scanning path, and driving the sensor array to scan according to the steering angle sequence to obtain the position coordinate, the time mark, the electromagnetic echo signal value and the environment depth value of each scanning point;
1044. And distributing a path weight value equal to the region characteristic enhancement value for each scanning point, integrating the information of all the scanning points, and forming a scanning data set containing the position coordinates, the time marks, the electromagnetic echo signal values, the environment depth values and the path weight values.
In the above scheme, the region scanning priority refers to a weight coefficient generated based on the feature enhancement value in the feature enhancement record, and is used for quantifying the emergency degree of different regions on the sensor scanning, and the high-priority region needs to be covered preferentially to improve the detection efficiency. The optimal scanning path refers to a sensor motion track generated by a path planning algorithm, and needs to cover all areas with high priority and areas with low signal intensity, minimize the total steering angle and ensure the scanning efficiency and the data integrity. The steering angle sequence is a sensor unit steering instruction sequence obtained by decomposing an optimal scanning path and comprises angle change direction and amplitude parameters, and the sensor is driven to execute multi-angle scanning according to a preset path. The path weight value refers to a weight coefficient dynamically allocated according to the characteristic enhancement value and the environment depth value of the area where the scanning point is located, and is used for identifying the importance level of the point data in the whole analysis. The scanning data set refers to an original data set collected by the sensor array according to an optimal path, and comprises structured data of position coordinates, time marks, electromagnetic echo signal values, environment depth values and path weight values, and supports subsequent feature analysis and algorithm optimization.
In the embodiment of the present application, firstly, in step 1041, based on the historical scan parameter optimization value stored in the feature enhancement record and the feature parameters such as the target distribution density and the collision strength of the corresponding region, the region scan priority matrix is generated by a dynamic weight distribution algorithm. The matrix carries out weighted fusion on the characteristic enhancement value, the electromagnetic echo intensity and the perceived data of the thermal radiation deviation, for example, a decision model is constructed by adopting a analytic hierarchy process, the characteristic enhancement value is used as a criterion layer weight, and the dynamic priority is generated by combining the electromagnetic signal attenuation degree. The high priority areas will be marked as forced scan targets, ensuring that resources are preferentially allocated to critical detection areas. For example, a region with a characteristic enhancement value of 0.8 or more and an electromagnetic intensity lower than 50% of the mean value is marked as a high priority region, and forced scanning is performed to ensure that resources are preferentially allocated to the region.
Next, step 1042 geometrically correlates the high priority region with the low signal strength region by spatial overlay analysis to construct an overlay constraint map. And generating a shortest path covering all constraint nodes by dynamic programming by adopting an improved traveling business problem solving algorithm and taking the highest priority area as a starting point. For example, use AThe algorithm performs local path optimization, and combines the steering angle penalty function to forcedly cover all high-priority and low-signal areas and simultaneously realize total steering angle minimization. For example, after the high priority region is associated with the signal attenuation region, the system plans a helical scan path with the total steering angle decreasing from 120 ° to 65 ° to ensure full coverage of the critical region.
Then, a steering angle sequence of the detector unit, for example, a step adjustment of 0 ° to 15 ° to 30 °, is generated based on the optimal scan path decomposition in step 1043, and the multi-degree-of-freedom mechanical arm is driven to move synchronously with the sensor array according to the steering angle sequence. In the scanning process, four-dimensional space-time data of each scanning point are collected, wherein the four-dimensional space-time data comprise position coordinates, time marks, electromagnetic echo signal values and environment depth values. For example, when the scanning path is decomposed into a steering command sequence of 0 DEG to 15 DEG to 30 DEG, the laser radar is driven to synchronously acquire data with the infrared sensor, and the position of an acquired scanning point is (X=120.5 mm, Y=30.2 mm, Z=5.1 mm), the time stamp is 2025-06-10T14:30:00.000Z, the electromagnetic intensity is-85 dB and the depth value is 5.12m.
Finally, in step 1044, a path weight value is allocated to each scan point, where the path weight value is calculated by weighting the feature enhancement value of the located area and the environmental depth value according to a preset ratio. Then, the system integrates the position coordinates, time marks, electromagnetic echo signal values, environment depth values and path weight values of the scanning points, constructs a structured data set containing space-time metadata, signal characteristics and weight identifiers, and supports subsequent machine learning model training and three-dimensional reconstruction algorithm processing. For example, a certain scan point is located in a high-priority corrosion area, the characteristic enhancement value is 0.85, the environmental depth is 1.1m, and the weight is calculated as. This point data is labeled as a high confidence sample for subsequent defect classification model training.
In practical application, a defect point is stored as. In a four-dimensional space, the electromagnetic intensity dip is detected through a first-order difference from 58dB to 35dB and the depth gradient mutation is detected from 10mm/m to 28mm/m, and a morphological closed operation polymerization abnormal region is combined and marked as a phi 8cm sector shielding region formed by shielding of a mechanical arm bracket. According to a grid dividing method, dividing a space into units of 0.05mX0.05mX0.05mX0.05munit, finding that the characteristic point density of a defect dense area of 0.3mX0.2machieves 1200 points/m 3, triggering a dynamic weight distribution algorithm to increase the electromagnetic sensor gain of a shielding area from 1.2 times to 1.8 times so as to penetrate metal interference, and adjusting a background area filter window from 7X 7 to 5X 5 to inhibit noise. The sampling frequency is optimized through a genetic algorithm, and the sampling interval of the shielding area is shortened from 20ms to 5ms so as to capture dynamic characteristics. The adjusted parameters are fused with new data through an incremental learning algorithm, the electromagnetic intensity in the four-dimensional feature space is restored to 52dB, the collision intensity is reduced from 0.85 to 0.35, and the feature enhancement record shows that the combination of the parameters enables the defect detection rate to be improved to 99.5%. The log record comprises a parameter adjustment sequence, a density change curve and a conflict comparison table, and provides a self-adaptive optimization basis for a subsequent algorithm.
The whole scheme of the 104 realizes intelligent scanning control of the sensor array by utilizing the optimized parameters in the characteristic enhancement record and combining the electromagnetic echo signal characteristics. The method comprises the steps of mapping a characteristic enhancement value into a region scanning priority, dynamically locking a key target region and a weak signal region, forcibly covering all high-value regions by a path optimization algorithm and minimizing steering loss to generate a high-efficiency scanning path, driving a sensor array to accurately acquire space-time synchronous multi-dimensional data along the path, and simultaneously endowing each scanning point with a path weight derived from the characteristic enhancement value. Finally, a multidimensional scanning data set fused with the target value weight and the physical signal is formed, the pertinence and the integrity of data acquisition in a complex scene are obviously improved, and a structured information substrate with value gradients is provided for subsequent analysis.
105. And establishing a target track prediction model based on the scanning data set and the historical track data, and performing parallel optimization on the target track prediction model by using distributed computing nodes so as to coordinate parameter synchronization and decision optimization of the unmanned intelligent turntable, thereby realizing reconstruction of multi-target continuous tracks in a shielding environment.
Alternatively, step 105 may specifically include the steps of:
1051. extracting position coordinates, time marks and path weight values from the scanning data set, forming a space-time sequence of the scanning data based on the position coordinates and the time marks, and converting the path weight values into track confidence weights;
1052. aligning a position sequence in the historical track data with a space-time sequence of the scanning data, and establishing a dynamic displacement equation based on the track confidence weight to complete the construction of a target track prediction model;
1053. after the target track prediction model is deployed to the distributed computing nodes, each node processes a target data subset and optimizes the target track prediction model by exchanging parameters of the dynamic displacement equation, so that the spatial deviation between a prediction position and scanning data is minimized;
1054. coordinating an unmanned intelligent turntable based on an optimization result to feed back the predicted position to a detector unit to adjust a scanning angle focus predicted area, and simultaneously comparing the track confidence weight with a set threshold value to adjust the signal transmitting power of the detector;
1055. And iteratively executing the prediction process of the target track prediction model and the parameter updating process of the dynamic displacement equation under the shielding environment, and generating a continuous track sequence by calculating a connection track break point through extrapolation of the dynamic displacement equation.
The step 1055 specifically includes the steps of recording a last effective position coordinate as an initial break point when track break is detected, recording a first recapture position coordinate as a final break point, using the initial break point as an initial position, predicting the position of the next time point by using the dynamic displacement equation, inserting the predicted position as a virtual scanning point into a scanning data set, giving a damping track confidence weight to form an extended data set, updating parameters of the dynamic displacement equation based on the extended data set, carrying out forward iterative prediction from the initial break point to an intermediate time point in time sequence, taking the position of the intermediate time point at the moment as a forward predicted position, carrying out reverse iterative prediction from the final break point in time sequence to the same intermediate time point, taking the position of the intermediate time point at the moment as a reverse predicted position, and connecting the initial break point, the intermediate connection point and the final break point to form a continuous track segment when the spatial deviation between the forward predicted position and the reverse predicted position is smaller than a tolerance threshold.
In the above scheme, the historical track data refers to a position sequence recorded by the moving object in time sequence in the space-time dimension, and comprises coordinate points and millisecond-level precision time stamps, which are used for analyzing the motion mode and the behavior rule. The track confidence weight refers to a quantization index calculated based on the reliability of track points, and the weight is dynamically adjusted based on the characteristics of the comprehensive speed, the repeatability, the roundtrip and the like, and is used for filtering noise or abnormal tracks. The target track prediction model is a prediction framework integrating physical dynamics and a machine learning algorithm, inputs historical tracks and environment semantics, and outputs a future position sequence. The continuous track sequence is to eliminate the complete space-time track after interruption and connect the virtual points of the shielding interval through extrapolation algorithm.
In the embodiment of the present application, step 1051 first extracts key features from original scan data, including three-dimensional coordinates of a target, a time stamp, and a region weight value. The position coordinates reflect the geometric position of the target, the time stamp records the data acquisition time, the region weight value is calculated through the signal intensity or the target density, and the region importance is marked. The system sorts the discrete data according to time to form a space-time sequence, and suppresses noise through a Kalman filtering algorithm, so that the position estimation accuracy is improved. For example, after the dense point cloud acquired by the laser radar is filtered, stable characteristic points are reserved and abnormal values are removed. Then, the path weight value is mapped into the track confidence weight of the 0-1 interval through the Sigmoid function, the high confidence area weight can reach 0.9, and the low confidence area is reduced to 0.3. This process provides high quality input for the subsequent predictive model, ensuring that the algorithm focuses on the critical areas.
Then, the historical track data and the current time-space sequence are aligned in a time synchronization way through a step 1052, and a cubic spline interpolation is adopted to fill in a time gap so as to ensure that the resolution ratio of the historical track data and the current time-space sequence is consistent. For example, the historical track is 10 frames per second, the scan data is 5 frames per second, and the intermediate frame data is generated by linear interpolation. Based on the aligned data, the system establishes a dynamic displacement equation, the equation fuses the uniform velocity model and the LSTM network prediction capability, and the area with high weight is given higher priority. For example, because of the high confidence weight, the displacement equation preferentially matches the historical curvature parameters to generate a predicted trajectory conforming to the physical laws in the industrial pipeline elbow region. Meanwhile, an adaptive learning rate mechanism is introduced to optimize equation parameters, and coefficients are adjusted through a gradient descent method, so that prediction errors are reduced.
The trajectory prediction model is then deployed to the distributed computing cluster, each node independently processing a subset of the target data, via step 1053. The node adopts an Adam optimizer to adjust the parameters of the dynamic displacement equation, and the gradient is summarized through a parameter server and the global model is updated. For example, the node A calculates the gradient of the displacement equation of a certain area, and then uploads the gradient to the server, and the server fuses all the gradients of the nodes and then issues an updated value. The communication protocol employs an asynchronous messaging mechanism that reduces latency and improves throughput. In the intelligent warehouse scene, when 10 AGVs run simultaneously, the distributed optimization reduces the path prediction delay by 40%, and the global consistency error is controlled within 5 cm.
The unmanned turret is then coordinated to adjust the detector scan angle based on the optimized predicted position, for example by a PID controller to raise the turret speed to 200 °/sec to align with the predicted area, via step 1054. The system compares the track confidence weight with a threshold, if the weight is lower than the threshold, the signal transmitting power is reduced, otherwise, the power is increased to a peak value. For example, when a high confidence target is detected in a security scene, the turntable rapidly turns to and activates a millimeter wave radar high power mode to realize the identification of a tiny target within 300 meters. The power adjustment combines low precision computing techniques to reduce energy consumption while maintaining detection accuracy.
Finally, a repair mechanism based on a dynamic displacement equation and bidirectional prediction is provided by step 1055 for the problem of track interruption caused by signal shielding or equipment interference in a complex environment. When the system detects that the track signal is lost or the position jump exceeds a preset threshold value, firstly, the last effective position before interruption is recorded as an initial interruption point, and the first effective position after signal recovery is captured as a termination interruption point. And taking the initial break point as a reference, extrapolating and predicting the position of the next time step through a dynamic displacement equation, generating a virtual scanning point, giving confidence weight attenuated with time, and inserting the virtual scanning point into the original data set to form an extended data set. Then, the system re-optimizes parameters of the dynamic displacement equation based on the extended data set, for example, the coefficient of the kinematic model is adjusted through a gradient descent method, and the prediction accuracy is improved. After parameter update, from the initial break pointAnd terminating the point of interruptionRespectively carrying out forward and reverse iterative prediction according to time sequence until reaching the same intermediate time point. Based on the optimized dynamic equation, fromStarting to calculate the position frame by frame to generate a forward track sequence. From the slaveReverse reckoning to generate a backward track sequence. Calculating the same intermediate time point of two sequencesTime positional deviation: If it is If the error is smaller than the tolerance threshold, the track is judged to be repairable. The process introduces an adaptive tolerance mechanism, dynamically adjusts the threshold according to the complexity of the environment, and interpolates the threshold by cubic splineAndAt the position ofAnd smoothly connecting the two parts to generate continuous track segments. For example, the interpolation function is: wherein the coefficients a, b, c, d are traced at both ends by The position, speed and acceleration of the position are calculated. Finally, the system merges the initial break point, the intermediate connection point and the termination break point into a complete track and updates the global track database. The step ensures that the repaired track accords with the physical motion rule through track smoothness verification.
In practical application, in an unmanned turntable bearing ball tracking scene, a system extracts data points with coordinates of x=0.325 meters, y=0.148 meters and z=0.082 meters and 0.7 path weights thereof at a position of 12.800 seconds based on a scanning data set, converts the data points into track confidence weights of 0.85, and establishes a dynamic displacement equation by combining a historical track to be optimized in parallel through distributed nodes. When it is detected that the bearing ball is trace-interrupted at 13.2 seconds of the time stamp, the last effective point P1 coordinate is recorded as x=0.332 meters, y=0.153 meters, z=0.085 meters, and the first effective point P2 coordinate is captured again at 13.6 seconds of the time stamp as x=0.348 meters, y=0.162 meters, z=0.092 meters. Forward predicted timestamp 13.4 second position with P1 as starting break pointReverse prediction of a time point position with P2 as a termination point of the break is performed with x=0.340 m, y=0.158 m, and z=0.088 mFor x=0.341 meters, y=0.157 meters, z=0.089 meters, the spatial deviation is calculated=0.0017 Meters less than the tolerance threshold of 0.005 meters, generating a continuous track segment by cubic spline interpolation, wherein the interpolation coefficients comprise a=0.008, b= -0.12, c=1.35, d=0.085. And synchronously coordinating the unmanned turntable to increase the laser radar scanning angular rate to 200 degrees per second to focus on a predicted area, and increasing the millimeter wave transmitting power to 30 dB and milliwatts to penetrate oil stain shielding. Through five iterative optimization, the track interruption repair rate reaches 95.3%, the target tracking delay is reduced from 120 milliseconds to 28 milliseconds, and the millimeter-level continuous reconstruction of the 12 ball motion tracks of the turntable is realized.
According to the overall scheme 105, a dynamic displacement driven target track prediction model is constructed by fusing the scanning data with the weights and the historical track information, and parallel optimization of the model is realized by using distributed computation. The system firstly converts path weight into track confidence weight, establishes a space-time associated prediction model, continuously reduces the space deviation between a predicted position and actual scanning through the cooperation of distributed nodes and optimizing model parameters, dynamically coordinates an unmanned turntable based on an optimization result to adjust the scanning focusing area and the transmitting power of a detector to form a perception-decision closed loop, and adopts a bidirectional iteration prediction and virtual track point insertion mechanism to generate a continuous track sequence through convergence verification of forward prediction and reverse prediction aiming at track interruption caused by shielding. Finally, the lossless reconstruction of the multi-target track is realized in a complex shielding environment, and the robustness and the continuity of track prediction are obviously improved.
The following is a complete example for steps 101-105, as shown in fig. 2, in the automobile parts warehouse, the unmanned intelligent turret is equipped with 77GHz millimeter wave radar and binocular stereoscopic depth camera, and the AGV aisle data is synchronously collected at 30Hz frequency. When the turntable monitors 6 cargo AGVs, the millimeter wave radar emits frequency modulation continuous waves at intervals of 33ms, strong electromagnetic echoes of the metal goods shelf and medium-intensity reflection of an AGV body are captured, and the depth camera generates 1920 multiplied by 1080 depth map through parallax calculation, wherein the accuracy reaches +/-2 mm. The system ensures that radar emission pulses are aligned with camera exposure strictly through a hardware synchronous trigger, and all data are added with microsecond-level timestamps generated by the FPGA. The analysis module splits the radar signal into discrete points of time stamp, azimuth angle and signal intensity, and the depth data is analyzed into time stamp, world coordinates and depth value. Through time sliding window matching, the radar point at the time of t= 1689152000.124000 is associated with the same-frame depth point, and finally a structured observation data set containing 200 ten thousand records is constructed, wherein each format is as followsThe space-time synchronization error is less than 0.1ms.
Then, based on the data set, the system separates multidimensional features, namely firstly, 76-78GHz band-pass filtering is applied to electromagnetic signals, then, thermal radiation features are extracted through 40dB segmentation of a dynamic threshold value, an AGV motor continuous >60dB overheat area is identified, canny edge detection is adopted to depth data to calculate local gradients, gradient threshold value >8cm/m separation surface texture features are set, a place of 22cm/m concave deformation gradient peak value of the surface of an AGV container is successfully captured, meanwhile, based on continuous 10-frame position data, motion vectors are calculated through improving a Lucas-Kanade algorithm, and the transverse speed 0.8 m/s+/-0.05 m/s and the rotation angular speed 15 DEG/s of the AGV03 are separated. Then building four-dimensional feature space, expanding the data structure intoFor example, the feature vector of the AGV02 at t= 1689152000.456 is. Finally, through combined anomaly detection, when the electromagnetic intensity of a certain AGV is suddenly reduced from 52dB to 28dB and the depth gradient is suddenly changed from 12cm/m to 35cm/m due to the shielding of a shelf, a shielding area with the size of 1.2m multiplied by 0.8m is marked by the system, wherein the difference of heat radiation of the area and the normal area is more than 24dB, and the difference of texture gradient is more than 18cm/m.
Then, aiming at the shielding area, the system quantifies the spatial characteristics, adopts a 0.25m < 3 > voxel grid to count the target distribution density, shows that the density of the tail area of the shielded AGV04 is only 12 points/m < 3 > and the density of the normal area is more than 35 points/m < 3 >, and combines the heat radiation intensity deviation and the texture gradient deviation when calculating the characteristic collision intensity, and the collision value is obtained by normalization weighting. The detector is dynamically adjusted according to an optimization rule, millimeter wave radar gain is increased from 25dB to 40dB, beam width is narrowed from 45 degrees to 20 degrees to focus a shielding area, and depth camera exposure time is increased from 20ms to 40ms and an HDR mode is started. Rescanning after the parameters are validated, the new data features are obviously improved, the electromagnetic intensity at the tail of the AGV04 is restored to 45dB, the depth gradient is increased to 12cm/m, the collision intensity is reduced to 0.32, and the whole process is recorded as a structural feature enhancement log.
Based on the optimal parameters in the feature enhancement record, the radar gain is 40 dB/beam 20 DEG, the camera exposure is 40ms, and the system calculates the optimal scanning path. Constructing an electromagnetic intensity weight graph, resetting the weight of a shielding recovery area to 2.0 and the weight of a normal area to 1.0, superposing a conflict thermodynamic diagram, adding an exploration factor of 1.5 in a high-history conflict area, and adopting an improvement AThe algorithm plans the path, and the cost function fuses the distance weight, the characteristic weight and the conflict factor. The total path length generated was 35.6m, containing 128 scan sites, with 82 points densely distributed in the occlusion region accounting for 64%. The cradle head is driven to perform multi-angle scanning at the angular speed of 15 degrees/s, the spiral progressive track radius is adopted in the original shielding area of the AGV04 from 0.5m to 0.2m, and each position stays for 100ms to collect multi-mode data. Finally, a scanning data set with weights is generated, the weight coefficient of each data additional path is between 0.5 and 2.0, for example, the critical point P45 data is marked asThe total data set is up to 15 ten thousand pieces, covering the complete surface features of 6 AGVs.
And finally, integrating the current scanning data set and 10 ten thousand pieces of track data in history to construct an LSTM-CNN hybrid prediction model, inputting a depth gradient map and a thermodynamic diagram by CNN branch processing spatial characteristics, extracting local textures by 3-layer convolution, inputting a speed vector and an angular speed sequence by LSTM branch processing time sequence characteristics, and memorizing states by 128 units. Through Spark distributed cluster parallel training, 1 Driver Node coordinates tasks, 3 Executor nodes respectively process model partitions, node1 optimizes CNN convolution kernel, node2 optimizes LSTM hidden layer, and Node3 fuses space-time characteristics. After 200 iterations, the model reduces the track prediction error to 0.15m on the test set. After deployment, the turret decisions are coordinated, when the AGV05 enters the historic high collision zone, the model triggers parameter adjustments 300ms in advance and generates the evasive path point sequence [ x=8.12, y=4.56 ] → [ x=8.35, y=4.78 ]. Finally, under the continuous shielding environment, the track reconstruction integrity of 6 AGVs reaches 99.2%, the average positioning error is 0.18m, and the accuracy is improved by 4.3 times compared with that of the traditional method.
Fig. 3 is a schematic structural diagram of a multi-target automatic tracking system based on an unmanned intelligent turntable according to an embodiment of the present application, as shown in fig. 3, the system includes:
the acquisition module 31 is configured to acquire electromagnetic echo signals and environmental depth data of a target area through a detector of the unmanned intelligent turntable, and perform synchronous calibration on the electromagnetic echo signals and the environmental depth data in a time dimension to form an observation data set;
The construction module 32 is configured to separate thermal radiation characteristics, surface texture characteristics and motion state characteristics of the target based on the physical characteristic differences of different sensors in the observation data set, construct a four-dimensional feature space including space-time coordinates, and identify an abnormal area of the surface of the target as a shielding area;
The adjusting module 33 is configured to adjust signal parameters of the detector according to the target distribution density in the four-dimensional feature space and the feature collision strength of the shielding area so as to enhance the target feature of the shielding area, and feed back the adjusted signal parameters to the four-dimensional feature space to form a feature enhancement record;
The scanning module 34 is used for driving the sensor array to perform multi-angle scanning, and calculating an optimal scanning path according to the signal parameter optimization value in the characteristic enhancement record and the acquired electromagnetic echo signal so as to generate a scanning data set with weight;
And the prediction module 35 is configured to establish a target track prediction model based on the scan data set and the historical track data, and perform parallel optimization on the target track prediction model by using distributed computing nodes to coordinate parameter synchronization and decision optimization of the unmanned intelligent turntable, so as to realize reconstruction of a multi-target continuous track in a shielding environment.
The multi-target automatic tracking system based on the unmanned intelligent turntable shown in fig. 3 may execute the multi-target automatic tracking method based on the unmanned intelligent turntable shown in the embodiment of fig. 1, and its implementation principle and technical effects are not repeated. The specific manner in which the various modules and units perform operations in the multi-target automated tracking system based on the unmanned intelligent turret in the above embodiments has been described in detail in the embodiments related to the method, and will not be described in detail herein.
In one possible design, an unmanned intelligent turret-based multi-objective automatic tracking system of the embodiment of FIG. 3 may be implemented as a computing device, as shown in FIG. 4, that may include a storage component 41 and a processing component 42;
The storage component 41 stores one or more computer instructions for execution by the processing component 42.
The processing component 42 is used in an unmanned intelligent turret-based multi-target automatic tracking method as described above with respect to the embodiment of fig. 1.
Wherein the processing component 42 may include one or more processors to execute computer instructions to perform all or part of the steps of the methods described above. Of course, the processing component may also be implemented as one or more Application Specific Integrated Circuits (ASICs), digital Signal Processors (DSPs), digital Signal Processing Devices (DSPDs), programmable Logic Devices (PLDs), field Programmable Gate Arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic elements for executing the methods described above.
The storage component 41 is configured to store various types of data to support operations at the terminal. The memory component may be implemented by any type or combination of volatile or nonvolatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disk.
Of course, the computing device may necessarily include other components as well, such as input/output interfaces, display components, communication components, and the like.
The input/output interface provides an interface between the processing component and a peripheral interface module, which may be an output device, an input device, etc.
The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, and the like.
The computing device may be a physical device or an elastic computing host provided by the cloud computing platform, and at this time, the computing device may be a cloud server, and the processing component, the storage component, and the like may be a base server resource rented or purchased from the cloud computing platform.
The embodiment of the application also provides a computer storage medium which stores a computer program, and the computer program can realize the multi-target automatic tracking method based on the unmanned intelligent turntable in the embodiment shown in the figure 1 when being executed by a computer.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present application, and not for limiting the same, and although the present application has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present application.
Claims (10)
1. The multi-target automatic tracking method based on the unmanned intelligent turntable is characterized by comprising the following steps of:
Acquiring electromagnetic echo signals and environment depth data of a target area through a detector of an unmanned intelligent turntable, and synchronously calibrating the electromagnetic echo signals and the environment depth data in a time dimension to form an observation data set;
Based on the physical characteristic difference of different sensors in the observation data set, separating out the heat radiation characteristics, the surface texture characteristics and the motion state characteristics of the target, constructing a four-dimensional characteristic space containing space-time coordinates, and simultaneously identifying the abnormal area of the surface of the target as a shielding area;
according to the target distribution density in the four-dimensional feature space and the feature conflict intensity of the shielding area, adjusting signal parameters of the detector to strengthen the target feature of the shielding area, and feeding the adjusted signal parameters back to the four-dimensional feature space to form a feature enhancement record;
Calculating an optimal scanning path according to the signal parameter optimization value in the characteristic enhancement record and the acquired electromagnetic echo signal, and driving a sensor array to perform multi-angle scanning according to the optimal scanning path so as to generate a scanning data set with weight;
And establishing a target track prediction model based on the scanning data set and the historical track data, and performing parallel optimization on the target track prediction model by using distributed computing nodes so as to coordinate parameter synchronization and decision optimization of the unmanned intelligent turntable, thereby realizing reconstruction of multi-target continuous tracks in a shielding environment.
2. The method of claim 1, wherein the establishing a target track prediction model based on the scan data set and the historical track data, and performing parallel optimization on the target track prediction model by using distributed computing nodes to coordinate parameter synchronization and decision optimization of the unmanned intelligent turntable, and realizing reconstruction of a multi-target continuous track in an occlusion environment comprises:
extracting position coordinates, time marks and path weight values from the scanning data set, forming a space-time sequence of the scanning data based on the position coordinates and the time marks, and converting the path weight values into track confidence weights;
aligning a position sequence in the historical track data with a space-time sequence of the scanning data, and establishing a dynamic displacement equation based on the track confidence weight to complete the construction of a target track prediction model;
After the target track prediction model is deployed to the distributed computing nodes, each node processes a target data subset and optimizes the target track prediction model by exchanging parameters of the dynamic displacement equation, so that the spatial deviation between a prediction position and scanning data is minimized;
Coordinating an unmanned intelligent turntable based on an optimization result to feed back the predicted position to a detector unit to adjust a scanning angle focus predicted area, and simultaneously comparing the track confidence weight with a set threshold value to adjust the signal transmitting power of the detector;
and iteratively executing the prediction process of the target track prediction model and the parameter updating process of the dynamic displacement equation under the shielding environment, and generating a continuous track sequence by calculating a connection track break point through extrapolation of the dynamic displacement equation.
3. The method of claim 2, wherein said calculating a connection trace break point by extrapolation of said dynamic displacement equation generates a continuous trace sequence comprising:
when the track interruption is detected, recording the last effective position coordinate as an initial interruption point and recording the first recapture position coordinate as a termination interruption point;
The initial break point is taken as an initial position, the dynamic displacement equation is applied to predict the position of the next time point, the predicted position is taken as a virtual scanning point to be inserted into a scanning data set, attenuation type track confidence weights are given to form an expansion data set, and parameters of the dynamic displacement equation are updated based on the expansion data set;
Based on a dynamic displacement equation after parameter updating, forward iterating and predicting the initial break point to an intermediate time point according to time sequence, taking the position of the intermediate time point at the moment as a forward predicting position, recording a forward track sequence generated by track coordinates in the forward iterating and predicting process, and simultaneously reversely iterating and predicting the same intermediate time point according to time sequence from the final break point, taking the position of the intermediate time point at the moment as a reverse predicting position, and recording a backward track sequence generated by the track coordinates in the reverse iterating and predicting process;
And when the spatial deviation of the forward predicted position and the backward predicted position is smaller than a tolerance threshold, smoothly connecting the forward track sequence and the backward track sequence at the middle time point through cubic spline interpolation to generate a continuous track segment.
4. The method of claim 1, wherein the computing an optimal scan path in combination with the acquired electromagnetic echo signals according to the optimized values of the signal parameters in the feature enhancement record and driving the sensor array to perform multi-angle scanning according to the optimal scan path to generate the weighted scan data set comprises:
generating a region scanning priority based on the characteristic enhancement value in the characteristic enhancement record, wherein a region with higher characteristic enhancement value corresponds to a higher scanning priority;
Identifying a low signal intensity region and a high priority region with the scanning priority larger than a preset threshold by combining the intensity distribution of the electromagnetic echo signals, taking the highest priority region as a path starting point, forcibly covering all the high priority region and the low signal intensity region, simultaneously realizing the minimization of the total steering angle, and generating an optimal scanning path;
Generating a steering angle sequence of the detector unit based on the optimal scanning path, and driving the sensor array to scan according to the steering angle sequence to obtain the position coordinate, the time mark, the electromagnetic echo signal value and the environment depth value of each scanning point;
And distributing a path weight value equal to the region characteristic enhancement value for each scanning point, integrating the information of all the scanning points, and forming a scanning data set containing the position coordinates, the time marks, the electromagnetic echo signal values, the environment depth values and the path weight values.
5. The method according to claim 1, wherein the separating the thermal radiation feature, the surface texture feature, and the motion state feature of the target based on the physical characteristic differences of different sensors in the observation dataset, constructing a four-dimensional feature space including space-time coordinates, and identifying an abnormal region of the surface of the target as a shielding region, comprises:
Extracting electromagnetic echo signal values and environment depth values and corresponding time marks and position coordinates based on physical characteristic differences of different sensors in the observed dataset;
based on the change mode of the electromagnetic echo signal value, the thermal radiation characteristic is separated through signal intensity threshold comparison, the surface texture characteristic is separated through adjacent depth value difference calculation based on the local fluctuation mode of the environment depth value, and the motion state characteristic is separated through position deviation calculation of continuous time points based on the change trend of the position coordinates and the time marks;
Respectively associating corresponding time marks and position coordinates according to the heat radiation characteristics, the surface texture characteristics and the motion state characteristics to construct a four-dimensional characteristic space;
In the four-dimensional feature space, an abnormal region of the target surface is identified through abrupt change detection of the electromagnetic echo signal value and the environment depth value, and when the surface texture feature or the heat radiation feature of the abnormal region has abnormal decline of the intensity value in a local region, the abnormal region is marked as a shielding region.
6. The method according to claim 1, wherein adjusting the signal parameters of the detector to enhance the target features of the occlusion region according to the target distribution density in the four-dimensional feature space and the feature collision intensity of the occlusion region, and feeding back the adjusted signal parameters to the four-dimensional feature space to form a feature enhancement record, comprises:
counting the number of the feature points in the unit space area according to the aggregation degree of the position coordinates in the four-dimensional feature space so as to calculate the target distribution density;
in the shielding region, respectively comparing intensity values of the surface texture features and the heat radiation features in a normal region and the shielding region based on intensity value deviation of the surface texture features and the heat radiation features so as to calculate corresponding feature collision intensity;
according to the target distribution density and the characteristic collision intensity, adjusting signal parameters of the detector through a signal parameter optimization rule, and enhancing target characteristics of the shielding area;
And applying the adjusted signal parameters to a detector, re-acquiring signals of the shielding area, updating the numerical values of the target distribution density and the characteristic collision intensity in the four-dimensional characteristic space, and recording the adjustment process of the signal parameters, the target distribution density and the characteristic collision intensity to form a characteristic enhancement record.
7. The method of claim 1, wherein the acquiring, by the detector of the unmanned intelligent turret, the electromagnetic echo signal and the environmental depth data of the target area, and the calibrating the electromagnetic echo signal and the environmental depth data in synchronization in a time dimension, forms an observation data set, comprises:
Setting a detector group comprising an electromagnetic sensor and a depth sensor on the unmanned intelligent turntable, wherein the electromagnetic sensor emits electromagnetic waves at fixed time intervals and receives reflected signals to generate electromagnetic echo signals, and the depth sensor acquires depth information of a target area as environmental depth data based on an optical principle;
Automatically attaching a time stamp uniformly managed by an internal clock of the unmanned intelligent turntable when the electromagnetic sensor and the depth sensor generate data, and decomposing the electromagnetic echo signal and the environmental depth data into discrete data points containing the time stamp, the position coordinates and corresponding signal intensity values;
and performing one-to-one pairing on the electromagnetic echo signals and discrete data points of the environmental depth data based on the time markers to form data units, and integrating the data units into a structured data set comprising the time markers, the position coordinates and corresponding signal intensity values as an observation data set.
8. A multi-target automatic tracking system based on an unmanned intelligent turntable, comprising:
The acquisition module is used for acquiring electromagnetic echo signals and environment depth data of a target area through a detector of the unmanned intelligent turntable, and synchronously calibrating the electromagnetic echo signals and the environment depth data in a time dimension to form an observation data set;
The construction module is used for separating out the thermal radiation characteristics, the surface texture characteristics and the motion state characteristics of the target based on the physical characteristic differences of different sensors in the observation data set, constructing a four-dimensional characteristic space containing space-time coordinates, and identifying the abnormal area of the surface of the target as a shielding area;
The adjusting module is used for adjusting signal parameters of the detector according to the target distribution density in the four-dimensional feature space and the feature conflict intensity of the shielding area so as to enhance the target feature of the shielding area, and feeding the adjusted signal parameters back to the four-dimensional feature space to form a feature enhancement record;
the scanning module is used for driving the sensor array to perform multi-angle scanning, calculating an optimal scanning path according to the signal parameter optimization value in the characteristic enhancement record and the acquired electromagnetic echo signal so as to generate a scanning data set with weight;
and the prediction module is used for establishing a target track prediction model based on the scanning data set and the historical track data, and carrying out parallel optimization on the target track prediction model by utilizing a distributed computing node so as to coordinate the parameter synchronization and decision optimization of the unmanned intelligent turntable and realize the reconstruction of the multi-target continuous track in the shielding environment.
9. The computing device is characterized by comprising a processing component and a storage component, wherein the storage component stores one or more computer instructions, and the one or more computer instructions are used for being invoked and executed by the processing component to realize the multi-target automatic tracking method based on the unmanned intelligent turntable according to any one of claims 1-7.
10. A computer storage medium, wherein a computer program is stored, and when the computer program is executed by a computer, the method for automatically tracking multiple targets based on the unmanned intelligent turntable according to any one of claims 1 to 7 is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510926166.4A CN120428215B (en) | 2025-07-07 | 2025-07-07 | A multi-target automatic tracking method and system based on unmanned intelligent turntable |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510926166.4A CN120428215B (en) | 2025-07-07 | 2025-07-07 | A multi-target automatic tracking method and system based on unmanned intelligent turntable |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN120428215A CN120428215A (en) | 2025-08-05 |
| CN120428215B true CN120428215B (en) | 2025-09-26 |
Family
ID=96565319
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510926166.4A Active CN120428215B (en) | 2025-07-07 | 2025-07-07 | A multi-target automatic tracking method and system based on unmanned intelligent turntable |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN120428215B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117197695A (en) * | 2023-09-14 | 2023-12-08 | 思翼科技（深圳）有限公司 | Unmanned aerial vehicle holder camera target identification tracking method and system based on artificial intelligence |
| CN117974712A (en) * | 2023-12-04 | 2024-05-03 | 河南科技大学 | Self-adaptive multi-target tracking method based on sub-track |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SG183690A1 (en) * | 2007-08-06 | 2012-09-27 | Trx Systems Inc | Locating, tracking, and/or monitoring personnel and/or assets both indoors and outdoors |
| CN118711149A (en) * | 2024-05-17 | 2024-09-27 | 北京凌云光子技术有限公司 | Target tracking method, device and electronic equipment for small photoelectric turntable of unmanned vehicle |
| CN119941789A (en) * | 2024-12-31 | 2025-05-06 | 电子科技大学长三角研究院(衢州) | A multi-target tracking method based on 3D depth information decoupling |
-
2025
- 2025-07-07 CN CN202510926166.4A patent/CN120428215B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117197695A (en) * | 2023-09-14 | 2023-12-08 | 思翼科技（深圳）有限公司 | Unmanned aerial vehicle holder camera target identification tracking method and system based on artificial intelligence |
| CN117974712A (en) * | 2023-12-04 | 2024-05-03 | 河南科技大学 | Self-adaptive multi-target tracking method based on sub-track |
Also Published As
| Publication number | Publication date |
|---|---|
| CN120428215A (en) | 2025-08-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US12092476B2 (en) | Method and apparatus for updating an environment map used by robots for self-localization | |
| CN119540183B (en) | Bridge damage detection method and system | |
| CN119575368B (en) | Unmanned aerial vehicle multi-target tracking method and system based on vision and millimeter wave radar information fusion | |
| CN118050734A (en) | Robot positioning method and system based on binocular vision and laser scanning fusion | |
| CN120010517B (en) | An AI-based cold chain warehouse drone obstacle avoidance and path planning inventory method and system | |
| CN119717828B (en) | An intelligent obstacle avoidance and path optimization method and system for AGV | |
| CN118276597A (en) | Intelligent obstacle detouring control method and device based on dynamic obstacle perception and unmanned aerial vehicle | |
| CN118605536A (en) | Autonomous obstacle avoidance method for mobile perception robot based on SLAM technology | |
| CN119919821B (en) | Safety analysis method and system applied to storage robot complex driving scene | |
| CN119780955B (en) | Autonomous navigation method based on laser and vision fusion and related equipment | |
| CN119671063B (en) | Intelligent park management platform based on three-dimensional visualization | |
| CN119001659B (en) | A multi-radar omnidirectional detection method for unmanned aerial vehicle inertial navigation system | |
| CN120428215B (en) | A multi-target automatic tracking method and system based on unmanned intelligent turntable | |
| CN119115930B (en) | Robot operation control method and system and robot | |
| CN119722713A (en) | A method for processing infrared images in specific areas | |
| CN119622317A (en) | Robot environment perception data processing method, system, device and medium | |
| CN119562352A (en) | Indoor mobile target positioning method and system based on machine learning | |
| CN118548870A (en) | A robot positioning and mapping method based on SLAM and infrared imaging technology | |
| CN117521422A (en) | Immersion-based team behavior simulation system and method | |
| CN117690200A (en) | BIM model-based autonomous inspection system for hospital logistics robot | |
| CN119704195B (en) | Method for improving performance of 3D vision guidance task of robot based on large model | |
| EP4060614B1 (en) | Method for determining noise statistics of object sensors | |
| CN120451726B (en) | A three-dimensional target detection method and device with multi-sensor time adaptive synchronization | |
| CN117636608B (en) | A method, device and medium for monitoring fire in a large space based on depth estimation | |
| CN117269951B (en) | Target tracking method with multi-view information enhancement in air and ground |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |