[go: up one dir, main page]

WO2024060209A1 - 一种处理点云的方法和雷达 - Google Patents

一种处理点云的方法和雷达 Download PDF

Info

Publication number
WO2024060209A1
WO2024060209A1 PCT/CN2022/120909 CN2022120909W WO2024060209A1 WO 2024060209 A1 WO2024060209 A1 WO 2024060209A1 CN 2022120909 W CN2022120909 W CN 2022120909W WO 2024060209 A1 WO2024060209 A1 WO 2024060209A1
Authority
WO
WIPO (PCT)
Prior art keywords
point
points
ground
height
point cloud
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.)
Ceased
Application number
PCT/CN2022/120909
Other languages
English (en)
French (fr)
Inventor
宋妍
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suteng Innovation Technologyco Ltd
Original Assignee
Suteng Innovation Technologyco Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Suteng Innovation Technologyco Ltd filed Critical Suteng Innovation Technologyco Ltd
Priority to PCT/CN2022/120909 priority Critical patent/WO2024060209A1/zh
Publication of WO2024060209A1 publication Critical patent/WO2024060209A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/88—Lidar systems specially adapted for specific applications
    • G01S17/89—Lidar systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection

Definitions

  • the present application relates to the field of computer technology, and in particular to a method and radar for processing point clouds.
  • LiDAR is an active remote sensing device that uses lasers as the light source and photoelectric detection technology. It is an advanced detection method that combines laser technology with modern photoelectric detection technology. LiDAR is widely used in autonomous driving, logistics vehicles, robots, vehicle-road collaboration, public smart transportation and other fields.
  • point cloud filtering can be used to remove background noise and retain effective point clouds.
  • this application provides a method and radar for processing point clouds, which can perform processing on ground points on the basis of removing abnormal noise points in the point cloud and meeting the sensing requirements of lidar. Effective identification ensures the accuracy of ground point identification.
  • a first aspect of this application provides a method for processing point clouds, including: obtaining a point cloud, which includes outlier points; determining ground points in the point cloud, where the ground points include first type points among outlier points, and The class point meets the height requirement, which is determined based on the ground point height of the previous frame point cloud of the current frame point cloud; at least the ground point is output.
  • the second aspect of the present application provides a device for processing point clouds, including: a point cloud acquisition module, a ground point determination module and a point cloud output module.
  • the point cloud acquisition module is used to obtain a point cloud, the point cloud includes outliers;
  • the ground point determination module is used to determine ground points in the point cloud, the ground points include first-class points among the outliers, the first-class points meet the height requirement, and the height requirement is determined based on the height of the ground points in the previous frame of the point cloud of the current frame;
  • the point cloud output module is used to output at least the ground points.
  • a third aspect of this application provides a board card, including the device for processing point clouds as mentioned above.
  • a fourth aspect of the present application provides a radar, including the above-mentioned device for processing point clouds.
  • a fifth aspect of the present application provides an electronic device, including: a processor; and a memory on which executable code is stored.
  • the processor is caused to execute the above method.
  • a sixth aspect of the present application provides a computer-readable storage medium on which executable code is stored.
  • the executable code is executed by a processor of an electronic device, the processor is caused to execute the above method.
  • a seventh aspect of this application provides a computer program product, which includes executable code. When the executable code is executed, the above method is implemented.
  • some embodiments of the present application determine the first type of points from the outlier points of the point cloud.
  • the first type of points meet the height requirements of the ground.
  • the height requirements are based on the upper limit of the current frame point cloud.
  • outliers may be abnormal noise points
  • the first type of points among the outlier points is retained to avoid misjudgment of ground points as abnormal noise points. This can effectively identify ground points and ensure the accuracy of ground point identification on the basis of removing abnormal noise points in the point cloud to meet the sensing requirements of lidar.
  • outlier points are screened from two perspectives: height consistency requirements and/or ground fitting straight line height range, which effectively improves the accuracy of judgment for the first type of points.
  • the outlier points are the second type points based on smoothness, and/or determine the outliers from the angle of the vector angle. Whether the point is a third type point. If the first type point is not a second type point and/or a third type point, the first type point can be deleted, which effectively improves the identification accuracy of ground points among outlier points.
  • whether the first type of point is a second type of point can be determined by the smoothness and difference between lines, further effectively improving the convenience and accuracy of identifying the second type of point.
  • the lidar since the lidar has a certain height relative to the ground, as the scanning distance increases, the distance between the two rows of points increases significantly. If a fixed smoothness threshold and/or difference is used degree threshold to determine point cloud smoothness, which may lead to inaccuracy. In this embodiment, the smoothness threshold and/or the difference threshold change as the distance between the point and the lidar changes, which further helps to improve the accuracy of the determined ground point.
  • the point cloud is divided into multiple fitting areas, and ground straight line fitting is performed respectively, which helps to reduce the computational complexity and reduce the consumption of computing resources.
  • the vector angle includes at least one of the first sub-angle, the second sub-angle, or the third sub-angle, so that whether the vector angle conforms to the ground can be judged from multiple angles. features, further helping to improve the accuracy of the determined ground points.
  • Figure 1 is a schematic diagram of a method for processing point clouds and an application scenario of radar according to an embodiment of the present application
  • Figure 2 is a schematic diagram of a vehicle-mounted lidar ground point cloud according to an embodiment of the present application
  • Figure 3 is a flow chart of a method for processing point clouds according to an embodiment of the present application.
  • Figure 4 is a schematic diagram of outlier points shown in an embodiment of the present application.
  • Figure 5 is a schematic diagram of the height of the fitted point cloud in the lidar coordinate system according to an embodiment of the present application.
  • Figure 6 is a schematic diagram showing an embodiment of the present application that does not meet the smoothness requirements
  • Figure 7 is a schematic diagram of the vector angle shown in an embodiment of the present application.
  • Figure 8 is another flow chart for processing point clouds according to an embodiment of the present application.
  • Figure 9 is another flow chart for processing point clouds according to an embodiment of the present application.
  • Figure 10 is a schematic diagram of an original point cloud according to an embodiment of the present application.
  • Figure 11 is a schematic diagram of a filtered point cloud according to an embodiment of the present application.
  • Figure 12 is a schematic diagram of a processed point cloud according to an embodiment of the present application.
  • Figure 13 is a schematic structural diagram of a device for processing point clouds according to an embodiment of the present application.
  • Figure 14 is a schematic structural diagram of a radar according to an embodiment of the present application.
  • FIG. 15 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
  • first, second, third, etc. may be used in this application to describe various information, the information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other.
  • first information may also be called second information, and similarly, the second information may also be called first information. Therefore, features defined as “first” and “second” may explicitly or implicitly include one or more of these features.
  • plurality means two or more than two, unless otherwise explicitly and specifically limited.
  • Vehicle-mounted radar The detection range is from 200 meters to 500 meters, and the identifiable physical attributes can include distance and reflectivity. It can be used for small machines such as vehicles and robots. Vehicle-mounted radar includes vehicle-mounted lidar, vehicle-mounted millimeter wave radar, vehicle-mounted ultrasonic radar, etc.
  • Lidar An active remote sensing device that uses a laser as a light source and adopts photoelectric detection technology. It is an advanced detection method that combines laser technology with modern photoelectric detection technology. It is composed of transmitting system, receiving system, scanning control system, data processing system and other parts. Its working principle is to transmit a detection signal to the target, and then process the received echo signal to obtain the target's distance, size, speed, reflectivity and other information. Its advantages are high resolution, high sensitivity, strong anti-interference ability, and not affected by dark conditions. It is widely used in fields such as autonomous driving, logistics vehicles, robots, vehicle-road collaboration, and public smart transportation.
  • Vehicle-mounted lidar By emitting outgoing light (such as a laser beam) with a wavelength of about 900nm, the outgoing light will be reflected by the obstacle when it encounters an obstacle.
  • the processing unit calculates the distance between the obstacle and the vehicle-mounted laser based on the time difference between the reflected light and the outgoing light. distance between radars. In addition, the processing unit can estimate the reflectivity of the target based on the cross-section of the reflected light.
  • Vehicle-mounted lidar has a high degree of integration due to its small size.
  • the system architecture suitable for autonomous driving scenarios can include mobile devices, networks and clouds.
  • Mobile devices include but are not limited to: cars, ships, robots, aircraft, etc.
  • the mobile device can be equipped with electronic devices such as sensors to obtain information about obstacles in the surrounding environment of the mobile device.
  • Electronic devices may include: radar, image sensors, etc.
  • Various electronic devices require the use of registers during operation or communication.
  • Embodiments of the present application provide a method and radar for processing point clouds to determine the first type of points that meet the height requirements of the ground from outlier points of the point cloud.
  • the first type of points are outliers, related technologies may misjudge them as abnormal noise points.
  • the first type of points that meet the height requirements among the outlier points can be retained. This can effectively identify ground points and ensure the accuracy of ground point identification on the basis of removing abnormal noise points in the point cloud to meet the sensing requirements of lidar.
  • Figure 1 is a schematic diagram of a method for processing point clouds and an application scenario of radar according to an embodiment of the present application.
  • FIG. 1 shows the hardware configuration of a vehicle 10 that supports assisted driving or automatic driving functions.
  • the vehicle 10 is equipped with at least one LiDAR (Light Detection and Ranging, LIDAR for short) 11 on the roof and/or the side of the body.
  • the detection area of LIDAR 11 can be fixed.
  • a certain LIDAR 11 can only be used to detect a preset area.
  • the detection area of LIDAR 11 can be adjustable.
  • the lidar on the car body can scan multiple detection areas by adjusting its posture, or it can scan multiple detection areas by adjusting the field of view of the lidar itself. .
  • the vehicle 10 can be equipped with 5 LIDARs 11: on the top of the vehicle, on the front side of the vehicle, on the rear side of the vehicle, on the left side of the vehicle and on the right side of the vehicle.
  • LIDARs 11 on the top of the vehicle, on the front side of the vehicle, on the rear side of the vehicle, on the left side of the vehicle and on the right side of the vehicle.
  • the vehicle 10 may also be equipped with a photographing device.
  • the shooting device can shoot the environment in front of the viewing angle at a prescribed viewing angle.
  • the photographing device may be a single-camera camera, a multi-camera camera, or the like.
  • the vehicle 10 may also be equipped with a plurality of millimeter wave radars surrounding the vehicle 10 .
  • the vehicle 10 is equipped with four millimeter wave radars, with the left side in front of the vehicle, the right side in front of the vehicle, the left side behind the vehicle, and the right side behind the vehicle as detection ranges.
  • Millimeter wave radar can detect the distance of an object present in each detection area and detect the relative speed of the object and the vehicle 10 .
  • the vehicle 10 can also be equipped with a positioning device 12, such as Beidou positioning device, Global Positioning System (GPS), etc.
  • the current position of the vehicle 10 can be determined via the positioning device 12 .
  • the vehicle 10 can also be equipped with an electronic control unit (Electronic Control Unit, ECU for short).
  • ECU Electronic Control Unit
  • the detection signal of at least one of the above-mentioned LIDAR 11, millimeter wave radar and positioning device 12 is sent to the ECU.
  • the ECU can detect and identify obstacles (such as roadblocks, moving objects, trees, adjacent vehicles, etc. around the vehicle 10 ) based on these signals.
  • the ECU can be physically divided into multiple units according to functions, which are collectively referred to as ECUs in this application.
  • the mobile device is described as a car, this description is not a limitation and is applicable to a variety of mobile devices, such as land robots, water robots, etc.
  • the point cloud processing method and radar in the embodiment of the present application can be applied to any one or more electronic devices that require the use of clocks, such as LIDAR 11, millimeter wave radar, positioning equipment 12, ECU or communication systems as shown in Figure 1 .
  • road signal control is coordinated to improve the quality and efficiency of road management.
  • the corresponding autonomous vehicle decisions can be determined based on the information sensed by the sensing system, and the safe distance between autonomous vehicles can be adjusted, so that the vehicles can drive on the road safely and reliably.
  • FIG2 is a schematic diagram of a ground point cloud of a vehicle-mounted laser radar according to an embodiment of the present application.
  • LiDAR will adopt a point cloud filtering scheme to remove background noise and retain effective point clouds.
  • related art point cloud filtering solutions can be based on neighborhood distance statistics to eliminate outliers. Effective points must meet the following conditions: the point cloud in the neighborhood must have a certain number of adjacent points, or the point cloud must be within The average distance of the neighborhood is within 1 Sigma, etc.
  • Ground point cloud plays a very important role in vehicle-mounted lidar perception.
  • the perception algorithm needs to be extracted from the ground and then clustered for perception of obstacles on the ground; or it can be used for curb and lane line detection for assisted driving. Therefore, the detection capability and integrity of ground point clouds are very important for vehicle-mounted lidar.
  • ground point clouds especially long-distance ground point clouds with small pitch angles, the distances between ground point clouds of adjacent scan lines are quite different. If the outlier elimination method of related technologies is still used for point cloud filtering, , it is easy to cause the ground point cloud to be mistakenly eliminated, thus affecting the perception ability of the lidar.
  • A, B and B, C are not equally spaced in Figure 2. Points O and P do not coincide. For example, when angle AOB and angle BOC are equal, the lengths of line segments AB and line segments BC are different. In addition, you can refer to the distance between different rows of point clouds shown in Figure 10.
  • ground point cloud is not specially processed, some ground point clouds will be judged as noise points and eliminated. If the point cloud filtering algorithm threshold is adjusted to ensure the integrity of the ground point cloud, some noise will be released, which will affect the judgment of application scenarios such as autonomous driving.
  • the first type of points that meet the height requirements are determined from the outlier points, and the probability that the first type of point is a ground point is determined larger.
  • Figure 3 is a flow chart of a method for processing point clouds according to an embodiment of the present application.
  • the method for processing a point cloud includes operations S310 to S330 .
  • a point cloud is obtained, and the point cloud includes outlier points.
  • the point cloud may be data collected by lidar.
  • LiDAR can be installed on a variety of mobile platforms, such as vehicles, robots, or exploration equipment. LiDAR obtains multiple reflection signals during the scanning process, and these reflection signals can be converted into point clouds.
  • Figure 4 is a schematic diagram of outlier points according to an embodiment of the present application.
  • the figure shows a partial image of a point cloud in a certain frame, in which most points are clustered together and some points are in an outlier state, that is, outlier points. These outliers may be mistakenly removed when filtering the point cloud.
  • outlier points may be points obtained through point cloud filtering processing.
  • Outliers refer to extremely large and small values in a time series that are far away from the general level of the series.
  • outliers can be identified from point clouds through statistics-based methods. Because outliers appear with low probability in the probability distribution model, low-probability data objects or data samples can be detected. However, the disadvantage is also obvious. Samples that appear with low probability are not necessarily outliers. If some ground points are outliers due to special ground structure, they can easily be misjudged as noise points.
  • Proximity-based methods identify outliers from point clouds and use the three nearest neighbors of the data object for modeling, so the points in the radius (R) area are significantly different from other object points in the data set.
  • the points in the radius (R) area are significantly different from other object points in the data set.
  • their second and third nearest neighbors are significantly farther than other objects (beyond a certain standard deviation, such as 1 sigma), so the objects in the R area can be marked as proximity-based Sexual outliers.
  • Clustering-based methods identify outliers from point clouds and detect outliers by examining the relationship between objects and clusters.
  • an outlier is an object that belongs to a small sparse cluster or does not belong to any cluster.
  • Classification-based methods identify outliers from point clouds. If there are class labels in the training data, it can be regarded as a classification problem. The idea of the problem is generally: train a classification that can distinguish "normal data" from outliers. Model. For example, you can use the model to construct a classifier that only describes the normal class, so that samples that do not belong to the normal class are outliers. Using only the normal class to detect outliers can detect new outliers that are not close to the outliers in the training set. . In this way, when a new outlier point comes in, as long as it is within the decision boundary of the normal class, it is a normal point, and if it is outside the decision boundary, it is an outlier point. For example, the construction of decision boundaries can refer to support vector machine (SVM).
  • SVM support vector machine
  • ground points in the point cloud are determined.
  • the ground points include first-type points among outlier points.
  • the first-type points meet height requirements.
  • the height requirements are ground points based on the previous frame point cloud of the current frame point cloud. Determined by height.
  • the height requirement can be dynamically changed.
  • the average height range of the ground point cloud in the point cloud of the previous frame of the current frame point cloud can be obtained, and the point cloud within the average height range is regarded as the height requirement.
  • the dynamically changing height requirements can automatically remove the inaccurate height requirements caused by the deviation of the fixed position of the lidar, and eliminate outliers that are actually ground points.
  • the height requirements are inaccurate due to changes in environmental conditions such as ground conditions, and outliers that are actually ground points are eliminated.
  • the height requirement can also be a preset height range.
  • the height range can be 0.9 meters to 1.1 meters.
  • the outlier points can be retained to improve the integrity of the ground point cloud.
  • ground points are output for subsequent processing of the point cloud to facilitate functions such as autonomous driving.
  • point clouds of vehicles, signs, traffic lights and other obstacles can also be output.
  • the processing method of ground point cloud features collected by vehicle-mounted lidar provided in this application can ensure the integrity of the ground point cloud while maintaining the denoising effect of point cloud filtering, thereby providing reliable input for the perception application of vehicle-mounted lidar. source.
  • the first category of points includes first subcategory points and/or second subcategory points, the first subcategory points meet height consistency requirements, and the second subcategory points are within the height range of the ground fitting straight line. .
  • the ground is usually relatively flat. If the height uniformity of each point in the point cloud is not good, the probability that it is a ground point is low. Therefore, there is a positive correlation between the high consistency of points in a point cloud and their probability of representing a ground point.
  • meeting the height requirements of the first type of point includes: the height of the first type of point is within the ground fitting straight line height range, and the ground fitting straight line height range is based on the previous frame of the current frame where the outlier point is located.
  • the range is determined by the ground height statistical results.
  • the ground height statistical results include at least one of the height average, variance and minimum value of the point cloud within the height limit range.
  • the first subcategory point is determined in the following manner: for each outlier point, if the height value of the outlier point is within the height range of the ground fitting straight line, then the outlier point is determined to be First subcategory point.
  • the second subclass point is determined as follows: for each of the outliers, a neighborhood point of the outlier is obtained, and if the height consistency between the outlier point and the neighborhood point is within a consistency range, the outlier point is determined to be a second subclass point.
  • the distance, height and reflectivity information of the point cloud data uploaded by the lidar device is used for processing.
  • Distance, height, and reflectivity information are all two-dimensional data. Traverse all pixels of the data, and take the data point with row number m and column number n as the target processing point. Its distance value is Dist(m,n), the height value is Height(m,n), and the reflectance value is Ref( m,n).
  • the radar installation height can be expressed as GroundHeightMeanPre.
  • the lowest point of height is represented as GroundHeightMinPre, and the value of GroundHeightMinPre is set to HeightTh1.
  • the height standard deviation within the height threshold range can be expressed as GroundHeightStdPre.
  • GroundHeightStdPre is set to 0.1 ⁇ GroundHeightMeanPre.
  • the average height within the height threshold range is expressed as GroundHeightMeanPre.
  • the ground point cloud feature preservation algorithm is processed, that is, it is further determined whether the outlier needs to be retained in order to use it as a ground point. Specifically, if the current point is judged to be an outlier, and the height of the current outlier point, Height(m,n), is greater than GroundHeightMeanPre-Q ⁇ GroundHeightStdPre, and less than GroundHeightMeanPre+Q ⁇ GroundHeightStdPre, and Height(m,n) is greater than GroundHeightMinPre -GroundHeightStdPre.
  • the outlier point can be initially considered to be a ground point and should be retained. Otherwise, it is considered not to be a ground point and the next point will be processed. It should be noted that after initially considering that the outlier point is a ground point, you can also determine whether the outlier point is a ground point by determining the height range in the next step.
  • Q is an integer greater than 1. For example, Q can be 2, 3, 4, etc. In one embodiment, the value of Q is 3, which represents three standard deviations (3-sigma rule). Using 3-sigma is an analysis method based on statistics for normally distributed data.
  • the first subcategory point can be determined in the following way. If the outlier point is the first type of point, that is, the outlier point meets the height range condition, take the Height(m-floor(L1/2) around the current point: m+floor(L1/2),n-floor(L2/2 ): The data points in the n+floor(L2/2)) neighborhood are analyzed as heights. If the absolute value of the difference between the height of each point in the neighborhood and Height(m,n) is less than the height difference threshold HeigthDiffTh and the number is greater than the statistical threshold HeightCntTh, the height consistency of the point is considered to be high. Otherwise, the high consistency of the outlier point is poor, and the processing of the next point will proceed.
  • the floor() function represents rounding down.
  • FIG. 5 is a schematic diagram of the height of the fitted point cloud in the lidar coordinate system according to an embodiment of the present application. Points with higher height consistency are included in the ground straight line fitting point set GroundPointList.
  • GroundPointList if expressed as a ground point cloud, it is approximately a straight line on the XOY-Z two-dimensional plane (i.e., Dist ⁇ cos ⁇ -Height plane, ⁇ is the pitch angle between the scanning line and the XOY plane), so it can be
  • the cos ⁇ -Height plane performs least squares fitting on the point set, and estimates the ground straight line to reduce the ground height range.
  • Equation (1) If the absolute value of the slope k is greater than kTh (kTh indicates that the slope is too large and does not conform to the ground characteristics), or r is less than rTh (rTh indicates that the data point set does not meet the threshold of a linear relationship), or the absolute value of the intercept b is greater than bTh ( bTh represents the threshold that the ground height is too high or too low), this point does not belong to the ground point. If the above three conditions are met, according to the fitting straight line equation, the current point height fitting value can be obtained as shown in Equation (1):
  • the ground height limit range can be updated to RefHeight-GroundHeightStdPre ⁇ -RefHeight+GroundHeightStdPre.
  • the point cloud has a large amount of data
  • the retained first type of points can be filtered based on some characteristics of the ground points.
  • the ground is usually relatively flat, and at least some of the first type points that do not conform to straight line fitting can be deleted based on straight line fitting.
  • the above method may also include the following operations.
  • the ground straight line fitting point set is fitted on a specific plane of a specific coordinate system to obtain at least one of the fitting straight line slope, fitting intercept and/or correlation coefficient.
  • the first type points that do not meet the ground fitting requirements are deleted to obtain updated first type points.
  • the data volume of the point cloud is very large, and the calculation of the entire point cloud is very computationally intensive.
  • the point cloud can be partitioned. After obtaining the updated first type of points, fitting can be performed based on methods such as least squares.
  • a particular coordinate system includes multiple fitting regions.
  • the ground straight line fitting point set is fitted on a specific plane of a specific coordinate system to obtain at least one of the fitting straight line slope, fitting intercept and/or correlation coefficient, including: for each fitting area , perform least squares fitting on the first type of points in the current fitting area, and obtain at least one of the slope of the fitting straight line, the fitting intercept and/or the correlation coefficient.
  • the ground fitting straight line height range can also be updated based on this to improve robustness. For example, an adaptive threshold based on data statistics is more robust than a preset fixed ground height range.
  • the above method may also include the following operations.
  • the height range is updated based on the fit line slope and the fit intercept.
  • the ground fitting straight line height range is updated based on the updated height range.
  • the ground straight line fitting point set GroundPointList is divided into N areas in the horizontal direction. Each area is ground fitted separately.
  • the fitting input point selects the current point and the accumulated points in front of the horizontal area where the current point is located.
  • the ground point set jumps by row to select M-1 points, and the jump step is Step.
  • Use the M points to perform least squares fitting of the ground straight line, and calculate the slope k, intercept b and correlation coefficient r of the fitted straight line. If the absolute value of the slope k is greater than kTh, or r is less than rTh, or the absolute value of the intercept b is greater than bTh, the point is not a ground point. If the conditions are met, fitting can be performed according to equation (1), and the ground fitting straight line height range is updated as:
  • the ground height is fitted based on the point cloud in the historical frame to obtain the fitted ground height range.
  • This makes it possible to determine whether the outlier points in the current frame are ground points based on the dynamically updated ground fitted straight line height range, effectively reducing The risk of misidentifying ground points as outlier noise points and deleting them.
  • the dynamically updated ground fitting straight line height range is an adaptive threshold based on data statistics, which is more robust.
  • the ground points also include at least one of the second type of points or the third type of points.
  • the second type of points meet the smoothness requirements
  • the third type of points meet the vector angle requirements.
  • the vector angle is a list of points. The angle between a line connecting adjacent points in the cloud and the ground plane.
  • This embodiment can also perform one-step analysis on the first type of points to improve the accuracy of ground point identification. Specifically, the first type of points obtained above can be further analyzed based on dimensions such as smoothness dimensions and/or vector angles.
  • the second type of points is determined as follows.
  • the inter-row smoothness and difference in the neighborhood of the current frame where the point is located are obtained.
  • the point cloud obtained after the laser radar scans the ground is distributed row by row.
  • the inter-row smoothness can refer to the point cloud in two adjacent rows of the outlier. As long as there is at least one row of point clouds that meets the smoothness requirement, the first category point can be retained.
  • the point is a second type point based on the inter-row smoothness and the smoothness threshold, and/or based on the difference and the difference threshold.
  • Figure 6 is a schematic diagram illustrating an embodiment of the present application that does not meet smoothness requirements. Referring to Figure 6, although the points in the point cloud are clustered and distributed, the smoothness is not high.
  • obtaining the inter-line smoothness and difference in the neighborhood of the current frame where the point is located may include the following operations.
  • a neighborhood point of the point in the current frame is obtained based on a preset window, and the ratio of the length and width of the preset window is a preset value.
  • the smoothness threshold and/or the dissimilarity threshold can be set based on the distance of the object relative to the lidar.
  • the smoothness threshold and/or the difference threshold can be set to change as the distance between the object represented by the point and the lidar changes.
  • the current outlier point if the current outlier point does not meet the height range, subsequent ground judgment is not performed and the next pixel point is processed. If the current outlier point meets the height range, take its surrounding neighborhood Dist(m-floor(L1/2): m+floor(L1/2), n-floor(L2/2): n+floor(L2/2 )) data points are analyzed as distances. Among them, L1 is the length of the vertical window, L2 is the length of the horizontal window, the typical size of L1 is 1, and the typical size of L2 is 2. Calculate the distance smoothness degree and difference degree of each row in the distance neighborhood. The calculation formulas of the distance smoothness degree SmoothCoef and the difference degree VarCoef can be as shown in Equation (2).
  • Judgment is based on the calculation result of (2). Although the ground points have large distance differences on the same scan line, the overall change is monotonous and smooth. It is judged whether it meets the following conditions: SmoothCoef is less than the threshold Th1, and VarCoef is less than the threshold. Th2, the threshold selection changes with distance, and the threshold is relatively relaxed when the distance is far. If the mth row where the current point is located satisfies the above conditions, and more than one row in two adjacent rows meets this condition, the point is considered to be a second type point, and is retained for the next pixel processing; otherwise, it can be judged Whether the first type point belongs to the third type point.
  • the vector angle includes at least one of a first sub-angle, a second sub-angle, or a third sub-angle.
  • the first sub-angle is between the first connection line and the corresponding ground plane.
  • the second sub-angle is the angle between the second connection line and the corresponding ground plane
  • the third sub-angle is the angle between the third connection line and the corresponding ground plane
  • the first connection line is the connection between the current point and the adjacent points in the same column of the current frame where the current point is located
  • the second connection is the connection between the current point and the next adjacent point in the same column of the current frame where the current point is located
  • the third connection is It is the line connecting the adjacent points in the same column of the current frame where the current point is located and the next adjacent point in the same column of the current frame where the current point is located.
  • FIG. 7 is a schematic diagram of vector angles according to an embodiment of the present application.
  • ⁇ is the vertical angular resolution of the radar. If the nth column where the current point is located satisfies the three angles that are less than the angle threshold AngleTh, and at least 2 columns in the adjacent columns satisfy the three angles that are less than the angle threshold, then the point is considered to be a ground point cloud, retain it, and proceed to the next pixel point processing; otherwise, the next pixel point is processed directly. All first-category points are traversed to determine whether they are third-category points, and then whether to retain the outlier point.
  • the point cloud may also include non-outlier points.
  • Figure 8 is another flowchart of processing point clouds according to an embodiment of the present application.
  • operation S310 may be replaced by operation S810 to obtain a point cloud, where the point cloud includes outlier points and non-outlier points.
  • Operation S330 may be replaced by operation S830, which outputs the ground points and at least part of the non-outlier points in the point cloud data.
  • operation S830 which outputs the ground points and at least part of the non-outlier points in the point cloud data.
  • point clouds of pedestrians, vehicles, traffic lights, street lights and other objects can also be output.
  • the point clouds of these objects can also be filtered, classified, etc., in order to implement functions such as autonomous driving.
  • non-outlier points are valid point clouds, including but not limited to pedestrians, plants, or man-made objects.
  • the height value of the current non-outlier point is greater than the height threshold HeightTh1 and less than the height threshold HeightTh2 (HeightTh is generally less than 0, the ground line is at a negative height position and can be set according to the vehicle radar installation height); and the distance is less than the distance threshold DistTh1, include this point into the ground point cloud height statistical range of the frame, count the lowest height point of the current frame, GroundHeightMinPre, and the height average value within the height threshold range, GroundHeightMeanPre, and the height standard deviation within the height threshold range, GroundHeightStdPre, used to correct the ground area of the next frame. selection, and then perform point cloud filtering of the next point.
  • GroundHeightMeanPre is set to the radar installation height
  • GroundHeightMinPre is set to HeightTh1
  • GroundHeightStdPre is set to 0.1*radar installation height.
  • FIG. 9 is another flowchart of processing a point cloud according to an embodiment of the present application.
  • the lidar uploads the collected point cloud data
  • the processor traverses all uploaded data points (m, n), performs point cloud filtering on the data pairs, and determines whether the current point is an outlier or a non-outlier. If it is not an outlier, the current point is determined to be a valid point, and the height average, variance and minimum value within the height range limited by the current frame are calculated and used to limit the height range of the next frame.
  • the current point is an outlier, then based on the ground height statistical value of the previous frame, determine whether the current point is within the ground height range (which can also be called the initial screening ground height range). If it is not within the ground height range, then Go over to the next point. If it is within the ground height range, select the height neighborhood data to determine whether the height consistency is high.
  • the ground height range which can also be called the initial screening ground height range.
  • the current point is classified into the ground point set to perform fitting point set, and the point set is divided into least squares fitting on the XOY-Z plane.
  • the ground fitting requirements it is judged whether the slope, intercept and correlation coefficient of the fitted straight line meet the ground fitting requirements. If they do not meet the ground fitting requirements, the next point is traversed. If the ground fitting requirements are met, the ground height limit range is updated based on the fitted straight line to improve robustness.
  • the smoothing burden and degree of difference meet the threshold requirements. If so, it is determined that the outlier point is a ground point and needs to be retained. If not, calculate the angle between the pairwise vectors of all row scan points in each column in the neighborhood and the horizontal plane. Next, determine whether the vector angle meets the threshold condition. If it does, it is determined that the current outlier point is a ground point and needs to be retained; if not, it is determined that the current outlier point is not a ground point but a noise point and needs to be eliminated.
  • the ground height is a condition that must be met for judgment, which may include at least one of height consistency and compliance with the ground fitting straight line height range. Smoothness or vector angle can satisfy one of them. Furthermore, the order of operations of the above operations can be adjusted.
  • Figure 10 is a schematic diagram of an original point cloud according to an embodiment of the present application.
  • the dotted arrows indicate ground line points
  • the single solid line arrows indicate lane line points
  • the double solid line arrows indicate noise points. It can be seen that the point cloud on the ground that is farther away from the lidar has a higher degree of dispersion, and the distance between two adjacent lines is also larger, and it is easy to be eliminated as outliers.
  • there are also outliers indicated by double solid arrows in the point cloud which are noise points and need to be removed.
  • Figure 11 is a schematic diagram of a filtered point cloud according to an embodiment of the present application. Please refer to Figure 10 and Figure 11 together. After point cloud filtering, the noise is removed. However, the point cloud of the ground far away from the lidar in Figure 10 was mistakenly deleted. Additionally, lane lines were deleted by mistake. This affects the expression of lidar's perception capabilities.
  • Figure 12 is a schematic diagram of a processed point cloud according to an embodiment of the present application.
  • the noise points are removed.
  • the point clouds on the ground farther away from the lidar are more completely preserved, and the lane lines are also preserved.
  • this embodiment can effectively remove abnormal noise in point clouds on the basis of satisfying the sensing capabilities of lidar.
  • Another aspect of the present application also provides a device for processing point clouds.
  • Figure 13 is a schematic structural diagram of a device for processing point clouds according to an embodiment of the present application.
  • the device 1300 for processing point clouds includes: a point cloud acquisition module 1310, a ground point determination module 1320, and a point cloud output module 1330.
  • the point cloud obtaining module 1310 is used to obtain a point cloud, which includes outlier points.
  • the ground point determination module 1320 is used to determine the ground points in the point cloud.
  • the ground points include the first type of points among the outlier points.
  • the first type of points meet the height requirements.
  • the height requirements are based on the previous frame point cloud of the current frame point cloud. determined by the height of the ground point.
  • the point cloud output module 1330 is used to output at least ground points. In addition, the point cloud output module 1330 can also be used to output non-outlier points.
  • the first type of points may include first sub-category points and/or second sub-category points, the first sub-category points meet the height consistency requirement, and the second sub-category points are within the height range of the ground fitting straight line.
  • Another aspect of the application also provides a radar.
  • Figure 14 is a schematic structural diagram of a radar according to an embodiment of the present application.
  • the radar 1400 may include circuitry.
  • a circuit could implement the method shown above for processing point clouds.
  • the circuit can be arranged on the circuit board 1410, and multiple chips, such as central control chips, can be arranged on the circuit board 1410.
  • the circuit board 1410 may be disposed in the housing 1420.
  • the radar may be a scanning radar or a non-scanning radar.
  • scanning laser radars include MEMS laser radars, mechanical laser radars, laser radars including multiple scanning devices, etc.
  • Non-scanning laser radars include flash laser radars, phased array laser radars, etc. This application does not limit the type of laser radar.
  • the MEMS solid-state lidar Take the MEMS solid-state lidar as an example. Since the MEMS solid-state lidar scans through the simple harmonic vibration of the galvanometer, its scanning path is realized in terms of spatial sequence. For example, it can be a slow axis from top to bottom and a fast axis from left. A scanning field of view reciprocating to the right. Therefore, the detection range of MEMS solid-state lidar is divided by dividing the field of view angle corresponding to the slow axis. For example, the slow axis of MEMS solid-state lidar corresponds to a vertical field of view angle of -13° to 13°.
  • the mechanical lidar in the scanning sensor Take the mechanical lidar in the scanning sensor as an example. Since the mechanical lidar drives the optical system through a mechanical drive device to rotate 360° to achieve scanning, a cylindrical detection area is formed with the lidar as the center. Therefore, the detection range corresponding to a mechanical lidar rotation of 360° is the detection range corresponding to the detection of one frame of data. Therefore, the detection range of a mechanical lidar cycle is generally divided by the degree of rotation.
  • the image is processed through the internal photosensitive component circuit and control component and converted into a digital signal that can be recognized by the computer. It is then input to the computer through a parallel port or USB connection, and the software then processes the image. reduction. Then the detection field of view of a period is generally divided by the area of the receiving detector.
  • Another aspect of the application also provides an electronic device.
  • FIG. 15 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
  • electronic device 1500 may include memory 1510 and processor 1520 .
  • the electronic device 1500 may also be provided with various sensors such as laser radar.
  • the processor 1520 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • a general-purpose processor may be a microprocessor or any conventional processor, etc.
  • Memory 1510 may include various types of storage units, such as system memory, read-only memory (ROM), and persistent storage. Among them, ROM can store static data or instructions required by the processor 1520 or other modules of the computer. Persistent storage may be readable and writable storage. Persistent storage may be a non-volatile storage device that does not lose stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (eg, magnetic or optical disk, flash memory) as the permanent storage device. In other embodiments, the permanent storage device may be a removable storage device (eg, floppy disk, optical drive).
  • System memory can be a read-write storage device or a volatile read-write storage device, such as dynamic random access memory.
  • System memory can store some or all of the instructions and data the processor needs to run.
  • memory 1510 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (eg, DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), magnetic disks, and/or optical disks may also be used.
  • memory 1510 may include a readable and/or writable removable storage device, such as a compact disc (CD), a read-only digital versatile disc (eg, DVD-ROM, dual-layer DVD-ROM), Read-only Blu-ray discs, ultra-density discs, flash memory cards (such as SD cards, min SD cards, Micro-SD cards, etc.), magnetic floppy disks, etc.
  • a readable and/or writable removable storage device such as a compact disc (CD), a read-only digital versatile disc (eg, DVD-ROM, dual-layer DVD-ROM), Read-only Blu-ray discs, ultra-density discs, flash memory cards (such as SD cards, min SD cards, Micro-SD cards, etc.), magnetic floppy disks, etc.
  • Computer-readable storage media do not contain carrier waves and transient electronic signals that are transmitted wirelessly or wired.
  • the memory 1510 stores executable code.
  • the processor 1520 can be caused to execute part or all of the above-mentioned methods.
  • the method according to the present application can also be implemented as a computer program or computer program product, which computer program or computer program product includes computer program code instructions for executing part or all of the steps in the above method of the present application.
  • the application may also be implemented as a computer-readable storage medium (or a non-transitory machine-readable storage medium or a machine-readable storage medium) with executable code (or computer program or computer instruction code) stored thereon,
  • executable code or computer program or computer instruction code
  • the processor of the electronic device or server, etc.
  • the processor is caused to execute part or all of the respective steps of the above method according to the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Electromagnetism (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • Optical Radar Systems And Details Thereof (AREA)

Abstract

一种处理点云的方法和雷达，处理点云的方法包括：获得点云，点云包括离群点（S310）；确定点云中的地面点，地面点包括离群点中的第一类点，第一类点满足高度要求，高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的（S320）；至少输出地面点（S330）。处理点云的方法能够在去除点云中的异常噪点，满足激光雷达的感知要求的基础上，对地面点进行有效识别，保证地面点识别的准确性。

Description

一种处理点云的方法和雷达 技术领域
本申请涉及计算机技术领域，尤其涉及一种处理点云的方法和雷达。
背景技术
激光雷达是一种用激光器作为发射光源，采用光电探测技术手段的主动遥感设备，是激光技术与现代光电探测技术结合的先进探测方式。激光雷达广泛应用于自动驾驶、物流车、机器人、车路协同、公共智慧交通等领域。
为了减少激光雷达输出点云中的异常噪点，可以通过点云滤波方式去除背景噪声，保留有效点云。然而，对地面点云进行滤波后，容易导致地面点云的误剔除，影响激光雷达的感知能力。
发明内容
为解决或部分地解决相关技术中存在的问题，本申请提供一种处理点云的方法和雷达，能够在去除点云中的异常噪点，满足激光雷达的感知要求的基础上，对地面点进行有效识别，保证地面点识别的准确性。
本申请第一方面提供一种处理点云的方法，包括：获得点云，点云包括离群点；确定点云中的地面点，地面点包括离群点中的第一类点，第一类点满足高度要求，高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的；至少输出地面点。
本申请第二方面提供一种处理点云的装置，包括：点云获得模块、地面点确定模块和点云输出模块。其中，点云获得模块用于获得点云，点云包括离群点；地面点确定模块用于确定点云中的地面点，地面点包括离群点中的第一类点，第一类点满足高度要求，高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的；点云输出模块用于至少输出地面点。
本申请第三方面提供一种板卡，包括如上述的处理点云的装置。
本申请第四方面提供一种雷达，包括如上述的处理点云的装置。
本申请第五方面提供一种电子设备，包括：处理器；以及存储器，其上存储有可执行代码，当可执行代码被处理器执行时，使处理器执行如上述的方法。
本申请第六方面提供一种计算机可读存储介质，其上存储有可执行代码，当可执 行代码被电子设备的处理器执行时，使处理器执行如上的方法。
本申请第七方面提供一种计算机程序产品，包括可执行代码，当可执行代码被执行时，实现如上的方法。
本申请提供的技术方案可以包括以下有益效果：
本申请实施例中，本申请的某些实施例，从点云的离群点中确定第一类点，该第一类点满足地面的高度要求，该高度要求是基于当前帧点云的上一帧点云来确定的高度范围。当离群点可能是异常噪点时，由于满足高度要求，为了满足激光雷达的感知能力，避免地面点被误判为异常噪点，保留离群点中的第一类点。这样能够在去除点云中的异常噪点满足激光雷达的感知要求的基础上，对地面点进行有效识别，保证地面点识别的准确性。
此外，本申请在某些实施例中，从高度一致性要求和/或地面拟合直线高度范围两个角度来对离群点进行筛选，有效提升了针对第一类点的判断精准度。
此外，本申请在某些实施例中，在得到第一类点之后，可以进一步从平滑度来判断离群点是否为第二类点，和/或，从向量夹角的角度来判断离群点是否为第三类点。如果该第一类点不是第二类点和/或第三类点，则可以删除该第一类点，有效提升了离群点中地面点的识别精准度。
此外，本申请在某些实施例中，可以通过行间平滑度和差异度来判断第一类点是否为第二类点，进一步有效提升识别第二类点的便捷度和准确度。
此外，本申请在某些实施例中，由于激光雷达相对于地面具有一定高度，随着扫描距离的增加，两行点之间的距离显著增大，如果采用固定的平滑度阈值和/或差异度阈值来判断点云平滑度，可能导致不准确。本实施例中，平滑度阈值和/或差异度阈值随着该点与激光雷达之间的距离改变而改变，进一步有助于提升确定的地面点的准确度。
此外，本申请在某些实施例中，将点云划分至多个拟合区中，分别进行地面直线拟合，有助于降低计算复杂度，降低对计算资源的消耗。
此外，本申请在某些实施例中，向量夹角包括第一子夹角、第二子夹角或者第三子夹角中至少一种，使得可以从多种角度判断向量夹角是否符合地面特征，进一步有助于提升确定的地面点的准确度。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，并不能限制本申请。
附图说明
通过结合附图对本申请示例性实施方式进行更详细地描述，本申请的上述以及其它目的、特征和优势将变得更加明显，其中，在本申请示例性实施方式中，相同的参考标号通常代表相同部件。
图1是本申请一实施例示出的处理点云的方法和雷达的一种应用场景的示意图；
图2是本申请一实施例示出的车载激光雷达地面点云的示意图；
图3是本申请一实施例示出的一种处理点云的方法的流程图；
图4是本申请一实施例示出的离群点的示意图；
图5是本申请一实施例示出的激光雷达坐标系下拟合点云高度示意图；
图6是本申请一实施例示出的不满足平滑度要求的示意图；
图7是本申请一实施例示出的向量夹角的示意图；
图8是本申请一实施例示出的处理点云的另一种流程图；
图9是本申请一实施例示出的处理点云的另一种流程图；
图10是本申请一实施例示出的原始点云的示意图；
图11是本申请一实施例示出的经滤波的点云的示意图；
图12是本申请一实施例示出的经处理的点云的示意图；
图13是本申请一实施例示出的处理点云的装置的结构示意图；
图14是本申请一实施例示出的一种雷达的结构示意图；
图15是本申请一实施例示出的电子设备的结构示意图。
具体实施方式
下面将参照附图更详细地描述本申请的实施方式。虽然附图中显示了本申请的实施方式，然而应该理解，可以以各种形式实现本申请而不应被这里阐述的实施方式所限制。相反，提供这些实施方式是为了使本申请更加透彻和完整，并且能够将本申请的范围完整地传达给本领域的技术人员。
在本申请使用的术语是仅仅出于描述特定实施例的目的，而非旨在限制本申请。在本申请和所附权利要求书中所使用的单数形式的“一种”和“该”也旨在包括多数形式，除非上下文清楚地表示其他含义。还应当理解，本文中使用的术语“和/或”是指并包含一个或多个相关联的列出项目的任何或所有可能组合。
应当理解，尽管在本申请可能采用术语“第一”、“第二”、“第三”等来描述各种信息，但这些信息不应限于这些术语。这些术语仅用来将同一类型的信息彼此区分开。例如，在不脱离本申请范围的情况下，第一信息也可以被称为第二信息，类似地，第二信息也可以被称为第一信息。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括一个或者更多个该特征。在本申请的描述中，“多个”的含义是两个或两个以上，除非另有明确具体的限定。
为了便于对本申请的理解，先对本申请涉及的部分概念进行说明。
车载雷达：探测距离如200米至500米，而且可识别的物理属性可以包括距离和反射率，可以用于车辆、机器人等小型机器。车载雷达包括车载激光雷达、车载毫米波雷达和车载超声波雷达等。
激光雷达：用激光器作为发射光源，采用光电探测技术手段的主动遥感设备，是激光技术与现代光电探测技术结合的先进探测方式。由发射系统、接收系统、扫描控制系统、数据处理系统等部分组成。其工作原理是向目标发射探测信号，然后将接收到回波信号进行处理，就可获得目标的距离、大小、速度、反射率等信息。其优点是分辨率高、灵敏度高、抗干扰能力强、不受黑暗条件影响等。广泛应用于自动驾驶、物流车、机器人、车路协同、公共智慧交通等领域。
车载激光雷达：通过发射如900nm左右波长的出射光(如激光束)，出射光遇到障碍物后会被障碍物反射，处理单元根据反射光和出射光之间的时间差计算障碍物与车载激光雷达之间的距离。此外，处理单元还可以根据反射光的横截面情况估算目标的反射率。车载激光雷达由于体积小，集成程度高。
在自动驾驶场景中，适用于自动驾驶场景的系统架构可以包括移动设备、网络和云端。移动设备包括但不限于：汽车、船舶、机器人、飞行器等。移动设备上可以设置有传感器等电子设备，以便获得移动设备周边环境中的障碍物信息等。电子设备可以包括：雷达、图像传感器等。多种电子设备在运行或通讯过程中需要使用寄存器。
本申请实施例提供一种处理点云的方法和雷达，从点云的离群点中确定满足地面的高度要求的第一类点。当第一类点是离群点时，相关技术可能将其误判为异常噪点。为了满足激光雷达的感知能力，避免地面点被误判为异常噪点，本申请实施例中可以保留离群点中满足高度要求的第一类点。这样能够在去除点云中的异常噪点满足激光雷达的感知要求的基础上，对地面点进行有效识别，保证地面点识别的准确性。
以下结合附图详细描述本申请实施例的技术方案。
图1是本申请一实施例示出的处理点云的方法和雷达的一种应用场景的示意图。
图1示出了支持辅助驾驶或者自动驾驶功能的车辆10的硬件构成。例如，车辆10的车顶和/或车身侧面搭载有至少一个激光雷达(Light Detection and Ranging，简称LIDAR)11。LIDAR 11的检测区域可以是固定的，如某个LIDAR 11可以仅用于对预设的某个区域进行检测。LIDAR 11的检测区域可以是可调的，如车身上的激光雷达可以通过调整姿态等方式对多个检测区域进行扫描，也可以通过调整激光雷达自身的视场角范围对多个检测区域进行扫描。具体地，车辆10可以搭载有5台LIDAR 11：车辆顶部、车辆前侧、车辆后侧、车辆左侧和车辆右侧。通过多台LIDAR 11，能够对车辆周围的区域内存在的物体的轮廓和到该物体的距离进行检测。
此外，车辆10上还可以搭载有拍摄装置。拍摄装置能够以规定的视角对视角的前方环境进行拍摄。例如，拍摄装置可以为单目相机、多目相机等。
另外，车辆10上还可以以围绕车辆10的方式搭载有多个毫米波雷达。例如，车辆10上搭载有4台毫米波雷达，以将车辆前方的左侧、车辆前方的右侧、车辆后方的左侧、车辆后方的右侧作为检测范围。通过毫米波雷达，能够检测各自的检测区域内存在的物体的距离，并且能检测该物体与车辆10的相对速度。
进一步地，车辆10上还可以搭载有定位设备12，如北斗定位设备、全球定位系统(Global Positioning System，简称GPS)等。通过定位设备12能够确定车辆10的当前位置。
此外，车辆10上还可以搭载有电子控制单元(Electronic Control Unit，简称ECU)。上述LIDAR 11、毫米波雷达和定位设备12中至少一种的检测信号发送给ECU。ECU能够基于这些信号对障碍物(如车辆10周边的路障、移动物体、树木、相邻车辆等)进行检测和识别。此外，ECU在物理上可以按功能分为多台，本申请将其统称为ECU。
需要说明的是，尽管可移动设备被描述为汽车，然而这样的描述并不是限制，多种移动设备都适用，如陆地机器人、水上机器人等。
本申请实施例的处理点云的方法和雷达可以应用于如图1所示的LIDAR 11、毫米波雷达、定位设备12、ECU或者通讯系统等需要使用时钟的任意一种或多种电子设备中。
在上述辅助驾驶、自动驾驶、智慧交通等应用场景中，快速、精确地感知可移动设备的周围环境是一个关键点。
在此以车辆的自动驾驶场景为例进行示例性说明。根据传感系统感知的车辆位置信息、障碍物信息和道路信息等，协调道路信号控制，从而提高道路管理质量和效率。具体地，可以根据传感系统感知的信息确定对应的自动驾驶车辆决策，调整自动驾驶车辆之间安全距离，这样便于实现车辆能够安全、可靠地在道路上行驶。
图2是本申请一实施例示出的车载激光雷达地面点云的示意图。
参见图2，以激光雷达坐标系XOY-Z为例，坐标系原点在地面的投影为P。激光雷达的回波强度会随探测距离的增加而衰减，由于环境噪声(如强光噪声、雨雪噪声等)、系统精度、电磁干扰等因素影响，会导致回波距离探测受到干扰，产生异常噪点，影响雷达的测距精度和测距能力。因此激光雷达都会采用点云滤波方案，去除背景噪声，保留有效点云。然而，相关技术的点云滤波方案可以是基于邻域距离统计的方法进行离群点剔除，有效点需满足如下条件：邻域内点云距离邻近的点满足一定的个数，或点云满足在邻域平均距离1西格玛(Sigma)之内等。
地面点云在车载激光雷达感知中有非常重要的作用，感知算法需要通过地面提取后再对地面上障碍物进行聚类感知；或者用于进行路沿和车道线检测，用于辅助驾驶。因此地面点云的探测能力和完整性对车载激光雷达来说非常重要。然而，对于地面点云，特别是俯仰角较小的远距离地面点云，相邻扫描线的地面点云间距离差异较大，如果仍然采用相关技术的离群点剔除的方法进行点云滤波，容易导致地面点云的误剔除，从而影响激光雷达的感知能力。具体参见图2中A、B和B、C两两之间不是等间距的。O点和P点不重合，如角AOB和角BOC相等时，导致线段AB和线段BC之间的长度不同。另外，可以参考图10所示的不同行点云之间的距离。
相关技术不易准确判断目标点是地面点，还是噪声点。如不对地面点云进行特殊处理，会导致部分地面点云被判定为噪声点而剔除。如果调整点云滤波算法阈值门限，保证地面点云的完整，那么又会有部分噪声被放出，均会影响自动驾驶等应用场景的判断。
在本申请的某些实施例中，针对车载激光雷达点云在滤波过程中丢失地面点的问题，从离群点中确定满足高度要求的第一类点，第一类点是地面点的概率较大。通过保留第一类点来降低在滤波过程中丢失的地面点，提升激光雷达的感知能力。
图3是本申请一实施例示出的一种处理点云的方法的流程图。
参见图3，该处理点云的方法包括操作S310～操作S330。
在操作S310，获得点云，点云包括离群点。
在本实施例中，点云可以是由激光雷达采集的数据。激光雷达可以安装在多种移动平台上，如车辆、机器人或者勘探设备上。激光雷达在扫描过程中得到多个反射信号，这些反射信号可以被转化为点云。
图4是本申请一实施例示出的离群点的示意图。
参见图4，图中示出了某一帧点云的局部图像，其中，大部分点聚集在一起，部分点处在离群状态，即离群点。这些离群点在对点云进行过滤时可能被误剔除。
在某些实施例中，离群点可以是通过点云滤波处理获得的点。
离群点是指一个时间序列中，远离序列的一般水平的极端大值和极端小值。
例如，可以通过基于统计的方法从点云中识别离群点。因为离群点在概率分布模型中低概率出现，可以通过检测低概率的数据对象或数据样本，不过缺点也较为明显，低概率出现的样本不一定也是离群点。如有地面点由于地面构造特殊，导致某个或某些地面点呈现离群状态，容易被误判为噪点。
基于邻近性的方法从点云中识别离群点，使用数据对象的三个最近邻来进行建模，那么半径(R)区域里面的显著不同于该数据集的其他对象点。对应R中的对象，它们的第二个第三个最近邻都显著比其他对象的更远(超出一定的标差，如1西格玛)，因此可以将R区域中的对象作一个标记为基于邻近性的离群点。
基于聚类的方法从点云中识别离群点，通过考察对象与簇之间的关系检测离群点。换而言之，离群点是一个对象，它属于小的稀疏簇或者不属于任何簇。
基于分类的方法从点云中识别离群点，如果训练数据中有类标号，则可以将其视为分类问题，该问题思路一般是：训练一个可以区分“正常数据”和离群点的分类模型。例如，可以使用模型构造一个仅仅描述正常类的分类器，这样不属于正常类的样本就是离群点，仅使用正常类检测离群点可以检测不靠近训练集中的离群点的新离群点。这样，当一个新离群点进来时，只要它位于正常类的决策边界内就为正常点，在决策边界外就为离群点。例如，决策边界的构建可以参考支持向量机(SVM)。
在操作S320，确定点云中的地面点，地面点包括离群点中的第一类点，第一类点满足高度要求，高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的。
在本实施例中，高度要求可以是动态变化的，如可以得到当前帧点云的上一帧点云中地面点云的平均高度范围，将点云处于该平均高度范围内作为高度要求。例如，当激光雷达的固定位置发生偏离时，动态变化的高度要求可以自动去除因激光雷达的固定位置发生偏离，导致的高度要求不准确，剔除了实为地面点的离群点。又例如， 当车辆处于急转弯、上坡或下坡时，由于地面状况等环境状况发生改变导致的高度要求不准确，剔除了实为地面点的离群点。
需要说明的是，高度要求也可以是预设的高度范围，如激光雷达距离地面1米高，则高度范围可以为0.9米～1.1米，当离群点的高度处于0.9米～1.1米时，则可以保留该离群点，以提升地面点云完整性。
在操作S330，至少输出地面点。
在本实施例中，通过输出地面点以便对点云进行后续处理，以便于实现诸如自动驾驶等功能。例如，除了需要输出地面点外，还可以输出车辆、标识牌、红绿灯等行车障碍物的点云。
本申请提供的车载激光雷达采集的地面点云特征的处理方法，可在维持点云滤波去噪效果的前提下，保证地面点云的完整性，从而为车载激光雷达的感知应用提供可靠的输入源。
以下对第一类点进行示例性说明。
在某些实施例中，第一类点包括第一子类点和/或第二子类点，第一子类点满足高度一致性要求，第二子类点在地面拟合直线高度范围内。其中，为了行车方便，地面通常是较平坦的，如果点云中各点的高度均匀性不好，则其是地面点的概率较低。因此，点云中各点的高度一致性与其表征地面点的概率之间正相关。
在某些实施例中，第一类点满足高度要求包括：第一类点的高度处于地面拟合直线高度范围内，地面拟合直线高度范围是根据该离群点所在当前帧的上一帧的地面高度统计结果来确定的范围，地面高度统计结果包括高度限定范围内的点云的高度平均值、方差和最小值中至少一种。
在某些实施例中，第一子类点通过如下方式确定：对于离群点中的每一个，如果该离群点的高度值在地面拟合直线高度范围内，则确定该离群点为第一子类点。
在某些实施例中，第二子类点通过如下方式确定：对于离群点中的每一个，获得该离群点的邻域点，如果该离群点和邻域点之间的高度一致性在一致性范围内，则确定该离群点为第二子类点。
在一个具体实施例中，使用激光雷达设备上传点云数据的距离、高度和反射率信息进行处理。距离、高度和反射率信息均为二维数据。遍历数据的所有像素点，以行号m，列号n的数据点为目标处理点，其距离值为Dist(m,n)，高度值为Height(m,n)，反射率值为Ref(m,n)。
在完成了对激光雷达点云进行滤波之后，可以通过如下操作确定点云中的离群点是否为地面点。以安装在汽车上的雷达为例，雷达安装高度可表示为GroundHeightMeanPre。高度最低点表示为GroundHeightMinPre，GroundHeightMinPre的值设为HeightTh1。高度阈值范围内高度标准差可以表示为GroundHeightStdPre，例如，GroundHeightStdPre设为0.1×GroundHeightMeanPre。高度阈值范围内高度平均值表示为GroundHeightMeanPre。
如果点云滤波后判断该点为离群点，则进行地面点云特征保持算法处理，即需要进一步确定是否需要保留该离群点，以便将其作为地面点。具体地，如果判断当前点为离群点，并且，当前离群点的高度Height(m,n)大于GroundHeightMeanPre-Q×GroundHeightStdPre，并且小于GroundHeightMeanPre+Q×GroundHeightStdPre，并且Height(m,n)大于GroundHeightMinPre-GroundHeightStdPre。如果满足该条件，认为符合地面高度范围，则可以初步认为该离群点是地面点，应该进行保留，否则认为不是地面点，进行下一个点的处理。需要说明的是，在初步认为该离群点是地面点之后，还可以通过下一步确定高度范围的判断，来确定离群点是否为地面点。上述计算公式中Q为大于1的整数，如Q可以为2,3,4等。在一个实施例中，Q的取值为3，表示三个标准差(3-sigma法则)，采用3-sigma是利用基于统计学上对正态分布的数据的一种分析方法。
关于第一子类点可以通过如下方式进行确定。如果离群点是第一类点，即离群点满足高度范围条件，取当前点周围Height(m-floor(L1/2)：m+floor(L1/2),n-floor(L2/2)：n+floor(L2/2))邻域的数据点作为高度分析。如果邻域内各点高度与Height(m,n)高度差绝对值小于高度差阈值HeigthDiffTh的个数大于统计阈值HeightCntTh，认为该点高度一致性较高。否则该离群点的高度一致性较差，进入下一个点的处理。floor()函数表示向下取整。
关于第二子类点可以通过如下方式进行确定。图5是本申请一实施例示出的激光雷达坐标系下拟合点云高度示意图。将高度一致性较高的点纳入地面直线拟合点集GroundPointList。参见图5，如表示为地面点云，在XOY-Z二维平面(即Dist˙cosθ-Height平面，θ为扫描线与XOY平面的俯仰夹角)上近似为一条直线，因此可以在Dist˙cosθ-Height平面对点集进行最小二乘拟合，估计地面直线用于缩小地面高度范围。例如，在进行地面拟合时，拟合输入点选取当前点以及从当前点的前面累积的地面点集按行跳跃选取M-1个点，跳跃步进Step(如Step＝1等)。使用该M个 点进行最小二乘拟合地面直线，计算拟合直线斜率k和截距b以及相关系数r。若斜率k的绝对值大于kTh(kTh表征坡度过大不符合地面特征的阈值)，或者r小于rTh(rTh表征数据点集不符合线性关系的阈值)，或者截距b的绝对值大于bTh(bTh表征地面高度过高或过低的阈值)，该点不属于地面点。如满足上述三个条件，根据拟合直线方程，可得到当前点高度拟合值如式(1)所示：
RefHeight＝k×Dist(m,n)+b   式(1)
相应地，可以更新地面高度限制范围为RefHeight-GroundHeightStdPre～-RefHeight+GroundHeightStdPre。
在某些实施例中，点云的数据量很大，可以基于地面点的一些特征对保留下来的第一类点进行筛选。例如，地面通常比较平坦，可以基于直线拟合的方式删除至少部分不符合直线拟合的第一类点。
具体地，上述方法还可以包括如下操作。
首先，将第一类点归入地面直线拟合点集。
然后，在特定坐标系的特定平面上对地面直线拟合点集进行拟合，得到拟合直线斜率、拟合截距和/或相关系数中至少一种。
接着，基于拟合直线斜率、拟合截距和/或相关系数中至少一种，删除不满足地面拟合要求的第一类点，得到更新第一类点。
在某些实施例中，点云的数据量很大，对整体点云进行计算的运算量很大，为了改善上述问题，可以对点云进行分区处理。在得到更新第一类点之后，可以基于最小二乘等方式进行拟合。
例如，特定坐标系包括多个拟合区。相应地，在特定坐标系的特定平面上对地面直线拟合点集进行拟合，得到拟合直线斜率、拟合截距和/或相关系数中至少一种，包括：对于每个拟合区，对当前拟合区中的第一类点进行最小二乘拟合，得到拟合直线斜率、拟合截距和/或相关系数中至少一种。
此外，在得到拟合直线斜率、拟合截距和/或相关系数中至少一种之后，还可以基于此更新地面拟合直线高度范围，以提升鲁棒性。例如，根据数据做统计的自适应阈值相对于预设的固定地面高度范围的鲁棒性更高。
具体地，上述方法还可以包括如下操作。
首先，基于拟合直线斜率和拟合截距拟合更新高度范围。
然后，基于更新高度范围更新地面拟合直线高度范围。
在一个具体实施例中，在水平方向将地面直线拟合点集GroundPointLi st分成N个区域，每个区域单独进行地面拟合，拟合输入点选取当前点以及从当前点所在水平区域前面累积的地面点集按行跳跃选取M-1个点，跳跃步进Step。使用该M个点进行最小二乘拟合地面直线，计算拟合直线斜率k和截距b以及相关系数r。若斜率k的绝对值大于kTh，或者r小于rTh，或者截距b的绝对值大于bTh，该点不是地面点。如满足条件，可按式(1)进行拟合，更新地面拟合直线高度范围为：
RefHeight-GroundHeightStdPre～-RefHeight+GroundHeightStdPre。
本实施例中基于历史帧中点云拟合地面高度，得到拟合地面高度范围，这样使得可以基于动态的更新地面拟合直线高度范围判断当前帧中离群点是否为地面点，有效减少了将地面点误判为离群点噪点进行删除的风险。此外，动态的更新地面拟合直线高度范围是根据数据做统计的自适应阈值，鲁棒性更高。
在某些实施例中，地面点还包括第二类点或者第三类点中至少一种，第二类点满足平滑度要求，第三类点满足向量夹角要求，向量夹角是一列点云中相邻点的连线与地平面之间的夹角。本实施例还可以对第一类点进行一步分析，以提升地面点识别的的准确度。具体地，可以基于平滑度维度和/或向量夹角等维度对上述得到的第一类点进行进一步分析。
在某些实施例中，第二类点通过如下方式确定。
首先，对于第一类点中的每个点，获得该点所在当前帧的邻域中行间平滑度和差异度。参见图10所示，激光雷达扫描地面后得到的点云是逐行分布的，行间平滑度可以指离群点的相邻两行中，只要有至少一行的点云满足平滑度要求，则可以保留该第一类点。
然后，基于行间平滑度和平滑度阈值，和/或，基于差异度和差异度阈值确定该点是否为第二类点。
图6是本申请一实施例示出的不满足平滑度要求的示意图。参见图6，点云中各点虽然聚集分布，但是，平滑度不高。
例如，获得该点所在当前帧的邻域中行间平滑度和差异度可以包括如下操作。
首先，基于预设窗口获得该点所在当前帧的邻域点，预设窗口的长度和宽度的比值为预设值。
然后，计算该点距离邻域点中每一列点的距离平滑程度和差异程度。
此外，由于激光雷达采集的一帧点云中不同行点云之间的间距是变化的，如距离 激光雷达越远的地面的点云，相邻行之间的行间距越大，具体参考图10所示。为了改善该问题，可以根据对象相对于激光雷达的距离来设置平滑度阈值和/或差异度阈值。
例如，可以设定平滑度阈值和/或差异度阈值随着该点表征的物体与激光雷达之间的距离改变而改变。
在一个具体实施例中，如果当前离群点不满足高度范围，不进行后续地面判断，进行下一个像素点处理。如果当前离群点满足高度范围，取其周围邻域Dist(m-floor(L1/2)：m+floor(L1/2),n-floor(L2/2)：n+floor(L2/2))的数据点作为距离分析。其中，L1为垂直方向窗口长度，L2为水平方向窗口长度，L1典型大小1，L2典型大小2。计算距离邻域内每一行的距离平滑程度和差异程度，距离平滑程度SmoothCoef、差异程度VarCoef的计算公式可以如式(2)所示。
Figure PCTCN2022120909-appb-000001
根据是(2)的计算结果进行判断，由于地面点在相同扫描行上虽然距离差异较大，但整体呈现单调变化且平滑变化趋势，判断其是否满足以下条件：SmoothCoef小于阈值Th1，VarCoef小于阈值Th2，阈值选取随距离变化，距离远则阈值相对放宽。如果当前点所在第m行满足上述条件，且相邻两行中有超过1行满足该条件，则认为该点为第二类点，将其保留，进行下一个像素点的处理；否则可以判断该第一类点是否属于第三类点。
在某些实施例中，向量夹角包括第一子夹角、第二子夹角或者第三子夹角中至少一种，第一子夹角是第一连线与对应的地平面之间的夹角，第二子夹角是第二连线与对应的地平面之间的夹角，第三子夹角是第三连线与对应的地平面之间的夹角，第一连线是当前点与该当前点所在当前帧的同列相邻点之间连线，第二连线是当前点与该当前点所在当前帧的同列的次相邻点之间连线，第三连线是当前点所在当前帧的同列的相邻点与该当前点所在当前帧的同列的次相邻点之间连线。
以下对第一子夹角、第二子夹角或者第三子夹角分别进行示例性说明。图7是本申请一实施例示出的向量夹角的示意图。
取当前点周围Height(m-floor(L1/2)：m+floor(L1/2),n-floor(L2/2)： n+floor(L2/2))和Dist(m-floor(L1/2)：m+floor(L1/2),n-floor(L2/2)：n+floor(L2/2))数据点做进一步分析。对于邻域内的每一列上的相邻点来说，以L1为1举例，参见图7，采用的激光雷达坐标系XOY-Z，示出了A、B、C、P1、P2、P3等点。需要满足向量AB、向量BC、向量AC与XOY水平面夹角ψ、Φ、θ分别小于角度阈值AngleTh，向量夹角可通过三角形公式求解。求解公式如式(3)～式(5)所示。
Figure PCTCN2022120909-appb-000002
BP1＝|HeightA-HeightB|       式(4)
Figure PCTCN2022120909-appb-000003
其中，ɑ为雷达的垂直角度分辨率。若当前点所在第n列满足三个角度均小于角度阈值AngleTh，并且相邻列中至少有2列满足三角度小于角度阈值，则认为该点为地面点云，将其保留，进行下一个像素点的处理；否则直接下一个像素点的处理。遍历完所有第一类点，以确定其是否为第三类点，进而确定是否保留该离群点。
在某些实施例中，点云还可以包括非离群点。图8是本申请一实施例示出的处理点云的另一种流程图。
参见图8，操作S310可以替换为操作S810，获得点云，点云包括离群点和非离群点。
操作S330可以替换为操作S830，输出地面点和点云数据中的至少部分非离群点。例如，除了需要输出地面点之外，还可以输出行人、车辆、红绿灯、路灯等对象的点云。需要说明的是，这些对象的点云同样可以经过了过滤、分类等处理，以便实现诸如自动驾驶功能。具体地，非离群点为有效点云，包括但不限于：行人、植物或者人造物等。
需要说明的是，若当前非离群点的高度值大于高度阈值HeightTh1，小于高度阈值HeightTh2(HeightTh一般小于0，地面线在负高度位置，可根据车载雷达安装高度设置)；且距离小于距离阈值DistTh1，将该点纳入该帧地面点云高度统计范围，统计当前帧的高度最低点GroundHeightMinPre以及高度阈值范围内高度平均值GroundHeightMeanPre和高度阈值范围内高度标准差GroundHeightStdPre，用于矫正下一帧地面区域的选取，然后进行下一个点的点云滤波处理。其中，对于激光雷达采集的第一帧点云，GroundHeightMeanPre设为雷达安装高度；GroundHeightMinPre设为HeightTh1；GroundHeightStdPre设为0.1*雷达安装高度。
图9是本申请一实施例示出的处理点云的另一种流程图。
参见图9，激光雷达将采集的点云数据上传，处理器遍历所有上传的数据点(m，n)，对数据对进行点云滤波处理，得到当前点是离群点还是非离群点。如果不是离群点，则确定当前点是有效点，统计当前帧限定高度范围内的高度平均值、方差和最小值，用于下一帧高度范围限定。
如果当前点是离群点，则根据上一帧的地面高度统计值，判断当前点是否处于地面高度范围内(也可以称为初筛地面高度范围内)，如果不处于地面高度范围内，则遍历下一点。如果处于地面高度范围内，则选取高度邻域数据，判断高度一致性是否较高。
如果高度一致性较低，则遍历下一点。如果高度一致性较高，则将当前点归入地面执行拟合点集，并分区在XOY-Z平面上对点集进行最小二乘拟合。
接着，判断拟合直线斜率、截距、相关系数是否满足地面拟合要求，如果不满足地面拟合要求，则遍历下一点。如果满足地面拟合要求，则根据拟合直线更新地面高度限制范围，提升鲁棒性。
然后，判断当前点高度是否满足更新后地面高度范围，如果不满足更新后地面高度范围，则遍历下一点。如果满足更新后地面高度范围，则计算距离邻域数据行间平滑程度和差异程度。
接着，判断平滑承担、差异程度是否满足阈值要求，如果是，则确定该离群点是地面点，需要保留。如果不是，则计算邻域中每一列中所有行扫描点两两向量与水平面的夹角大小。接着，判断向量夹角是否满足阈值条件，如果满足，则确定当前离群点是地面点，需要保留；如果不满足，则确定当前离群点不是地面点，是噪点，需要剔除。
重复以上操作，直至点云中所有点被遍历完成。需要说明的是，上述实施例中部分操作不是必要操作，虽然可能会影响结果的精确度，但是可以通过减少操作数量来提升计算速度、响应速度等。例如，地面高度是判别必须满足条件，其可以包括高度一致性和符合地面拟合直线高度范围中至少一种。平滑度或者向量夹角满足其一即可。此外，上述操作的操作顺序可以调整。
以下对点云处理效果进行示例性说明。
图10是本申请一实施例示出的原始点云的示意图。参见图10，虚线箭头地面线点，单实线箭头车道线点，双实线箭头噪点。可以看到，距离激光雷达越远的地面的 点云，离散度越高，并且相邻两行之间的间距也越大，容易被作为离群点而剔除。此外，点云中也存在双实线箭头所示的离群点，其为噪点，需要被去除。
图11是本申请一实施例示出的经滤波的点云的示意图。请一并参见图10和图11，经过点云滤波后，噪点被剔除。但是，图10中的距离激光雷达较远的地面的点云被误删除。此外，车道线也被误删除。这影响了对激光雷达的感知能力的表达。
图12是本申请一实施例示出的经处理的点云的示意图。参见图12，经过本申请实施例对离群点进行处理后，噪点被去除。但是，距离激光雷达较远的地面的点云较完整的保留了下来，并且车道线也被保留了下来。综上，本实施例能够在满足激光雷达的感知能力的基础上，有效去除点云中的异常噪点。
本申请的另一方面还提供了一种处理点云的装置。
图13是本申请一实施例示出的处理点云的装置的结构示意图。
参见图13，该处理点云的装置1300包括：点云获得模块1310、地面点确定模块1320和点云输出模块1330。
点云获得模块1310用于获得点云，点云包括离群点。
地面点确定模块1320用于确定点云中的地面点，地面点包括离群点中的第一类点，第一类点满足高度要求，高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的。
点云输出模块1330用于至少输出地面点。此外，点云输出模块1330还可以用于输出非离群点。
其中，第一类点可以包括第一子类点和/或第二子类点，第一子类点满足高度一致性要求，第二子类点在地面拟合直线高度范围内。
关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述，此处将不再做详细阐述说明。
本申请的另一方面还提供了一种雷达。
图14是本申请一实施例示出的一种雷达的结构示意图。
参见图14，该雷达1400可以包括电路。例如，电路可以实现如上所示的处理点云的方法。电路可以设置在电路板1410上，电路板1410上可以设置有多个芯片，如中控芯片等。电路板1410可以设置在壳体1420中。
雷达可以为扫描型雷达或者非扫描型雷达。其中，扫描型激光雷达包括MEMS型激光雷达，机械式激光雷达，包括多个扫描装置的激光雷达等。非扫描型激光雷达包 括Flash激光雷达、相控阵激光雷达等。本申请对于激光雷达的类型不作限制。
以MEMS固态激光雷达为例，由于MEMS固态激光雷达是通过振镜的简谐振动进行扫描的，其扫描路径从空间顺序上来说实现的例如可以是一个慢轴从上到下，快轴从左到右往复的一个扫描视场。因此，对于MEMS固态激光雷达的探测范围的划分通过是对慢轴对应的视场角进行划分。例如，MEMS固态激光雷达的慢轴对应的垂直视场角为-13°到13°。
以扫描型传感器中的机械式激光雷达为例，由于机械式激光雷达是通过机械驱动装置带动光学系统进行360°旋转实现扫描的，以激光雷达为圆心的一个圆柱形探测区域。因此，机械式激光雷达旋转360°对应的探测范围为探测一帧数据对应的探测范围，所以对机械式激光雷达一个周期探测范围的划分一般以旋转度数的划分。
而对于非扫描型激光雷达，是通过内部的感光组件电路及控制组件对图像进行处理并转换成电脑所能识别的数字信号，然后借由并行端口或USB连接输入到电脑后由软件再进行图像还原。则一个周期的探测视场一般以接收探测器的面积划分。
本申请的另一方面还提供了一种电子设备。
图15是本申请实施例示出的电子设备的结构示意图。
参见图15，电子设备1500可以包括存储器1510和处理器1520。此外，电子设备1500上还可以设置有激光雷达等多种传感器。
处理器1520可以是中央处理单元(Central Processing Unit，CPU)，还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器1510可以包括各种类型的存储单元，例如系统内存、只读存储器(ROM)和永久存储装置。其中，ROM可以存储处理器1520或者计算机的其他模块需要的静态数据或者指令。永久存储装置可以是可读写的存储装置。永久存储装置可以是即使计算机断电后也不会失去存储的指令和数据的非易失性存储设备。在一些实施方式中，永久性存储装置采用大容量存储装置(例如磁或光盘、闪存)作为永久存储装置。另外一些实施方式中，永久性存储装置可以是可移除的存储设备(例如软盘、光驱)。系统内存可以是可读写存储设备或者易失性可读写存储设备，例如动态随机访问内存。 系统内存可以存储一些或者所有处理器在运行时需要的指令和数据。此外，存储器1510可以包括任意计算机可读存储媒介的组合，包括各种类型的半导体存储芯片(例如DRAM，SRAM，SDRAM，闪存，可编程只读存储器)，磁盘和/或光盘也可以采用。在一些实施方式中，存储器1510可以包括可读和/或写的可移除的存储设备，例如激光唱片(CD)、只读数字多功能光盘(例如DVD-ROM，双层DVD-ROM)、只读蓝光光盘、超密度光盘、闪存卡(例如SD卡、min SD卡、Micro-SD卡等)、磁性软盘等。计算机可读存储媒介不包含载波和通过无线或有线传输的瞬间电子信号。
存储器1510上存储有可执行代码，当可执行代码被处理器1520处理时，可以使处理器1520执行上文述及的方法中的部分或全部。
此外，根据本申请的方法还可以实现为一种计算机程序或计算机程序产品，该计算机程序或计算机程序产品包括用于执行本申请的上述方法中部分或全部步骤的计算机程序代码指令。
或者，本申请还可以实施为一种计算机可读存储介质(或非暂时性机器可读存储介质或机器可读存储介质)，其上存储有可执行代码(或计算机程序或计算机指令代码)，当可执行代码(或计算机程序或计算机指令代码)被电子设备(或服务器等)的处理器执行时，使处理器执行根据本申请的上述方法的各个步骤的部分或全部。
以上已经描述了本申请的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术的改进，或者使本技术领域的其他普通技术人员能理解本文披露的各实施例。

Claims (17)

  1. 一种处理点云的方法，其特征在于，包括：
    获得点云，所述点云包括离群点；
    确定所述点云中的地面点，所述地面点包括所述离群点中的第一类点，所述第一类点满足高度要求，所述高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的；
    至少输出所述地面点。
  2. 根据权利要求1所述的方法，其特征在于，所述第一类点包括第一子类点和/或第二子类点，所述第一子类点满足高度一致性要求，所述第二子类点在地面拟合直线高度范围内。
  3. 根据权利要求1所述的方法，其特征在于，所述第一类点满足高度要求包括：所述第一类点的高度处于地面拟合直线高度范围内，所述地面拟合直线高度范围是根据该离群点所在当前帧的上一帧的地面高度统计结果来确定的范围，所述地面高度统计结果包括高度限定范围内的点云的高度平均值、方差和最小值中至少一种。
  4. 根据权利要求2所述的方法，其特征在于：
    所述第一子类点通过如下方式确定：对于所述离群点中的每一个，如果该离群点的高度值在地面拟合直线高度范围内，则确定该离群点为第一子类点；和/或
    所述第二子类点通过如下方式确定：对于所述离群点中的每一个，获得该离群点的邻域点，如果该离群点和所述邻域点之间的高度一致性在一致性范围内，则确定该离群点为第二子类点。
  5. 根据权利要求1所述的方法，其特征在于，还包括：
    将所述第一类点归入地面直线拟合点集；
    在特定坐标系的特定平面上对所述地面直线拟合点集进行拟合，得到拟合直线斜率、拟合截距和/或相关系数中至少一种；
    基于所述拟合直线斜率、所述拟合截距和/或所述相关系数中至少一种，删除不满足地面拟合要求的第一类点，得到更新第一类点。
  6. 根据权利要求5所述的方法，其特征在于，所述特定坐标系包括多个拟合区；
    所述在特定坐标系的特定平面上对所述地面直线拟合点集进行拟合，得到拟合直线斜率、拟合截距和/或相关系数中至少一种，包括：
    对于每个拟合区，对当前拟合区中的第一类点进行最小二乘拟合，得到所述拟合直线斜率、所述拟合截距和/或所述相关系数中至少一种。
  7. 根据权利要求5所述的方法，其特征在于，还包括：
    基于所述拟合直线斜率和所述拟合截距拟合更新高度范围；
    基于所述更新高度范围更新所述地面拟合直线高度范围。
  8. 根据权利要求1所述的方法，其特征在于，所述地面点还包括第二类点或者第三类点中至少一种，所述第二类点满足平滑度要求，所述第三类点满足向量夹角要求，所述向量夹角是一列点云中相邻点的连线与地平面之间的夹角。
  9. 根据权利要求8所述的方法，其特征在于，所述第二类点通过如下方式确定：
    对于第一类点中的每个点，获得该点所在当前帧的邻域中行间平滑度和差异度；
    基于所述行间平滑度和平滑度阈值，和/或，基于所述差异度和差异度阈值确定该点是否为第二类点。
  10. 根据权利要求9所述的方法，其特征在于，所述获得该点所在当前帧的邻域中行间平滑度和差异度，包括：
    基于预设窗口获得该点所在当前帧的邻域点，所述预设窗口的长度和宽度的比值为预设值；
    计算该点距离所述邻域点中每一列点的距离平滑程度和差异程度。
  11. 根据权利要求9所述的方法，其特征在于，所述平滑度阈值和/或所述差异度阈值随着该点表征的物体与激光雷达之间的距离改变而改变。
  12. 根据权利要求8至11任一项所述的方法，其特征在于，所述向量夹角包括第一子夹角、第二子夹角或者第三子夹角中至少一种，所述第一子夹角是第一连线与对应的地平面之间的夹角，所述第二子夹角是第二连线与对应的地平面之间的夹角，所述第三子夹角是第三连线与对应的地平面之间的夹角，所述第一连线是当前点与该当前点所在当前帧的同列相邻点之间连线，所述第二连线是所述当前点与该当前点所在当前帧的同列的次相邻点之间连线，所述第三连线是所述当前点所在当前帧的同列的相邻点与该当前点所在当前帧的同列的次相邻点之间连线。
  13. 根据权利要求1至11任一项所述的方法，其特征在于，所述离群点为通过点云滤波处理获得的点。
  14. 根据权利要求1至11任一项所述的方法，其特征在于，所述点云还包括非离群点；
    所述至少输出所述地面点包括：输出所述地面点和所述点云中的至少部分所述非离群点。
  15. 一种处理点云的装置，其特征在于，包括：
    点云获得模块，用于获得点云，所述点云包括离群点；
    地面点确定模块，用于确定所述点云中的地面点，所述地面点包括所述离群点中的第一类点，所述第一类点满足高度要求，所述高度要求是基于当前帧点云的上一帧点云的地面点高度来确定的；
    点云输出模块，用于至少输出所述地面点。
  16. 一种雷达，包括如权利要求15所述的处理点云的装置。
  17. 一种电子设备，其特征在于，包括：
    处理器；以及
    存储器，其上存储有可执行代码，当所述可执行代码被所述处理器执行时，使所述处理器执行如权利要求1至14任一项所述的方法。
PCT/CN2022/120909 2022-09-23 2022-09-23 一种处理点云的方法和雷达 Ceased WO2024060209A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/120909 WO2024060209A1 (zh) 2022-09-23 2022-09-23 一种处理点云的方法和雷达

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/120909 WO2024060209A1 (zh) 2022-09-23 2022-09-23 一种处理点云的方法和雷达

Publications (1)

Publication Number Publication Date
WO2024060209A1 true WO2024060209A1 (zh) 2024-03-28

Family

ID=90453716

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/120909 Ceased WO2024060209A1 (zh) 2022-09-23 2022-09-23 一种处理点云的方法和雷达

Country Status (1)

Country Link
WO (1) WO2024060209A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118470331A (zh) * 2024-07-12 2024-08-09 山东大学 基于自适应同心圆模型的地面点云分割方法及系统
CN120949184A (zh) * 2025-08-01 2025-11-14 北京建筑大学 一种地基雷达城市桥梁长时序监测的静杂波去除方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150248577A1 (en) * 2012-09-21 2015-09-03 Umwelt (Australia) Pty. Limited On-ground or near-ground discrete object detection method and system
WO2020006765A1 (zh) * 2018-07-06 2020-01-09 深圳前海达闼云端智能科技有限公司 地面检测方法、相关装置及计算机可读存储介质
CN111598780A (zh) * 2020-05-14 2020-08-28 山东科技大学 一种适用于机载LiDAR点云的地形自适应插值滤波方法
US20200372670A1 (en) * 2019-05-24 2020-11-26 Toyota Research Institute, Inc. Systems and methods for object detection including z-domain and range-domain analysis
CN112906449A (zh) * 2020-12-02 2021-06-04 北京中科慧眼科技有限公司 基于稠密视差图的路面坑洼检测方法、系统和设备
CN114170149A (zh) * 2021-11-17 2022-03-11 东南大学 一种基于激光点云的道路几何信息提取方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150248577A1 (en) * 2012-09-21 2015-09-03 Umwelt (Australia) Pty. Limited On-ground or near-ground discrete object detection method and system
WO2020006765A1 (zh) * 2018-07-06 2020-01-09 深圳前海达闼云端智能科技有限公司 地面检测方法、相关装置及计算机可读存储介质
US20200372670A1 (en) * 2019-05-24 2020-11-26 Toyota Research Institute, Inc. Systems and methods for object detection including z-domain and range-domain analysis
CN111598780A (zh) * 2020-05-14 2020-08-28 山东科技大学 一种适用于机载LiDAR点云的地形自适应插值滤波方法
CN112906449A (zh) * 2020-12-02 2021-06-04 北京中科慧眼科技有限公司 基于稠密视差图的路面坑洼检测方法、系统和设备
CN114170149A (zh) * 2021-11-17 2022-03-11 东南大学 一种基于激光点云的道路几何信息提取方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118470331A (zh) * 2024-07-12 2024-08-09 山东大学 基于自适应同心圆模型的地面点云分割方法及系统
CN120949184A (zh) * 2025-08-01 2025-11-14 北京建筑大学 一种地基雷达城市桥梁长时序监测的静杂波去除方法

Similar Documents

Publication Publication Date Title
CN111753609B (zh) 一种目标识别的方法、装置及摄像机
US11295521B2 (en) Ground map generation
CN109283538B (zh) 一种基于视觉和激光传感器数据融合的海上目标大小检测方法
CN112513679B (zh) 一种目标识别的方法和装置
WO2021097618A1 (zh) 点云分割方法、系统及计算机存储介质
WO2020243962A1 (zh) 物体检测方法、电子设备和可移动平台
WO2021012987A1 (zh) 物体检测方法、装置、系统及设备
CN114821526A (zh) 基于4d毫米波雷达点云的障碍物三维边框检测方法
CN114187579A (zh) 自动驾驶的目标检测方法、装置和计算机可读存储介质
CN115273062A (zh) 一种融合三维激光雷达和单目相机的3d目标检测方法
WO2022198637A1 (zh) 点云滤噪方法、系统和可移动平台
CN115639536B (zh) 基于多传感器融合的无人船感知目标检测方法及装置
WO2020253764A1 (zh) 确定行驶区域信息的方法及装置
WO2024060209A1 (zh) 一种处理点云的方法和雷达
WO2024159623A1 (zh) 一种浮空障碍物检测方法、装置和电子设备、存储介质
CN114002708B (zh) 一种面向无人艇应用的尾浪滤除方法
CN115994934B (zh) 数据时间对齐方法、装置以及域控制器
US20240385318A1 (en) Machine-learning based object detection and localization using ultrasonic sensor data
CN113902043B (zh) 目标识别方法、装置及设备
CN119156548A (zh) 一种点云评估方法及装置
CN112766100A (zh) 一种基于关键点的3d目标检测方法
CN116246162A (zh) 一种基于多传感器的无人船过闸识别方法
CN117409393A (zh) 一种焦炉机车激光点云与视觉融合障碍物检测方法及系统
CN116125440A (zh) 用于确定激光雷达传感器的最大作用范围的方法以及设备
CN115527034A (zh) 一种车端点云动静分割方法、装置及介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22959216

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22959216

Country of ref document: EP

Kind code of ref document: A1