HK40077647B - Method for precise 3d lidar aided satellite positioning - Google Patents

Description

Three-dimensional lidar-assisted high-precision satellite positioning method

Technical Field

This disclosure primarily relates to the field of autonomous driving or other types of autonomous systems for intelligent transportation systems. Specifically, this disclosure relates to a 3D LiDAR-assisted global navigation satellite system and method for NLOS detection and correction, which can improve positioning performance.

Background Technology

Autonomous driving is widely recognized as a remedy for excessive traffic congestion and anticipated accidents. However, the insufficient positioning accuracy of current solutions is one of the key issues hindering the arrival of autonomous driving in urban scenarios. With the increasing demand for ADV (Autonomous Driving Vehicles), positioning in urban environments becomes crucial.

GNSS, such as GPS and/or other similar satellite-based positioning technologies, is currently one of the primary means of providing global reference positioning for ADV (Advanced Vehicle Positioning) in intelligent transportation systems. With the increasing availability of multiple satellite constellations, GNSS can provide satisfactory performance over open sky areas. However, GNSS performance degrades significantly when much of the sky is obscured, posing a challenging problem. This situation is known as the "urban canyon" scenario. Typically, in highly urbanized cities, positioning accuracy decreases significantly due to signal reflections caused by static buildings and dynamic objects.

Specifically, GNSS-RTK is used for high-precision aerial mapping and positioning of Level 4 fully automated vehicles. Typically, GNSS-RTK positioning involves two steps: (1) estimating a floating-point solution based on received GNSS measurements; and (2) using a least-squares algorithm (e.g., LAMBDA) to solve for integer ambiguities based on the derived floating-point solution as an initial guess. With a fixed solution, centimeter-level positioning accuracy can be achieved based on dual-differential carrier and coded measurements over open areas. Unfortunately, the accuracy of GNSS-RTK is significantly reduced in urban canyons due to NLOS and multipath reception caused by GNSS signal reflection and obstruction by surrounding buildings. In practice, the significant decrease in GNSS-RTK positioning accuracy in urban canyons is mainly due to the occurrence of GNSS NLOS reception. Due to NLOS reception, the performance of GNSS positioning is highly affected by real-time environmental features such as buildings and moving objects. Some received GNSS signals are heavily contaminated, including large noise levels. According to previous research by the inventors, in highly urbanized areas, most received GNSS signals can be multipath or NLOS receptions. Therefore, the accuracy of floating-point solution estimation based on differential carrier and coded measurements is reduced, making it difficult to obtain a fixed solution for ambiguity.

Furthermore, the number of available satellites in urban canyons is limited due to signal obstruction from surrounding buildings. This distorts the geometry of satellite distribution, resulting in a large DOP (Depth of Opportunity). Consequently, the poor satellite geometry leads to significant ambiguity in the search space, making it difficult to obtain a fixed solution. In short, urban canyon scenarios introduce additional challenges to both steps of GNSS-RTK positioning.

Therefore, effectively sensing and understanding the surrounding environment is crucial for improving GNSS positioning in urban areas, as GNSS positioning heavily relies on sky visibility. The most well-known method for addressing GNSS NLOS reception is 3DMA GNSS positioning, such as NLOS exclusion and shadow matching based on 3D map construction information. However, these 3DMA GNSS methods have drawbacks: 1) they rely on the availability of 3D building models and initial guessing of the GNSS receiver's location; 2) they cannot mitigate NLOS reception caused by dynamic objects in the surrounding environment. Recent advances in 3DMA GNSS positioning methods are reviewed in detail in the inventors' previous work.

In the inventors' recent publication, a 3D LiDAR sensor, referred to as the "eyes" of an ADV (Advanced Driver Assistance Vehicle), a typical and indispensable onboard sensor for autonomous vehicles, has been used to detect NLOS caused by dynamic objects. Due to the limited field of view (FOV) of 3D LiDAR (-30° to +10°), only a portion of a double-decker bus can be scanned. Furthermore, the method heavily relies on the accuracy of object detection. Nevertheless, this is the first work to employ real-time object detection to assist GNSS positioning. A method was explored to detect surrounding buildings using real-time 3D LiDAR point clouds, rather than just detecting dynamic objects. Because of the limited field of view of 3D LiDAR, only a portion of buildings can be scanned. Therefore, information about building height is needed to detect NLOS reception caused by buildings. Instead of excluding detected NLOS reception, the inventors explored alternatives to correcting NLOS pseudorange measurements with the aid of LiDAR. The 3D LiDAR can measure the distance from the GNSS receiver to the surface of a building that may have reflected GNSS signals. The corrected and remaining healthy GNSS measurements can then be used for further GNSS positioning. Improved performance was achieved after correcting for detected NLOS satellites. However, the performance of this method depends on the accuracy of building and reflector detection. Both building and reflector detection may fail when building surfaces are highly irregular. The limited FOV of LiDAR remains a drawback for detecting dynamic objects and buildings. Overall, previous work has demonstrated the feasibility of using real-time airborne sensing (real-time point cloud) to detect GNSS NLOS. To overcome the limited FOV of 3D LiDAR, the inventors explored using a fisheye camera and a 3D LiDAR to detect and correct NLOS signals. The fisheye camera is used to detect non-line-of-sight signals. Simultaneously, the 3D LiDAR is used to measure the distance between the GNSS receiver and the potential reflector causing the NLOS signal. However, this method suffers from the problem of NLOS detection being sensitive to ambient lighting conditions.

Therefore, there is a need in the art for improved positioning methods and systems, particularly for autonomous driving systems capable of achieving high-precision positioning in very deep urban canyons. Furthermore, other desirable features and characteristics will become apparent from the following detailed description and appended claims, taken in conjunction with the accompanying drawings and background information.

Summary of the Invention

This paper presents a 3D LiDAR-assisted GNSS NLOS mitigation method and a system for implementing the method. The purpose of this disclosure is to provide a method for mitigating NLOS caused by static buildings and dynamic objects.

A first aspect of this disclosure provides a method for supporting vehicle positioning using a satellite positioning system. The method includes: receiving LiDAR factors and IMU factors from a 3D LiDAR sensor and an IIO; optimizing the integrated LiDAR factors and IMU factors using a local factor map to estimate the relative motion between two epochs; generating a 3D PCM as an auxiliary landmark satellite to provide low-elevation auxiliary landmark satellites; receiving GNSS measurements from the satellites via a GNSS receiver; detecting GNSS NLOS receptions from the GNSS measurements using the 3D PCM; and excluding the GNSS NLOS receptions from the GNSS measurements to obtain surviving GNSS satellite measurements, thereby improving the quality of GNSS measurements used for positioning of autonomous vehicles.

In one embodiment, the method further includes: performing GNSS-RTK floating-point estimation on the surviving GNSS satellite measurements to obtain a floating-point solution; performing ambiguity solving to obtain a fixed ambiguity solution; and determining a fixed GNSS-RTK positioning solution from the floating-point solution and the fixed ambiguity solution.

In one embodiment, the ambiguity solution is performed by applying the LAMBDA algorithm.

In one embodiment, the method further includes: feeding back the fixed GNSS-RTK positioning solution to the 3D LiDAR sensor and the LIO; and performing PCM correction using the fixed GNSS-RTK positioning solution to correct the drift of the 3D point cloud.

In one embodiment, the method further includes: using a least squares algorithm on the GNSS measurements to obtain an initial guess of the floating-point solution.

A second aspect of this disclosure provides a method for supporting vehicle positioning using a satellite positioning system. The method includes: generating a sliding window map (SWM) in real time based on 3D point clouds from a 3D LiDAR sensor and an AHRS, wherein the SWM provides an environmental description for detecting and correcting NLOS reception; accumulating 3D point clouds from previous frames into the SWM to enhance the FOV of the 3D LiDAR sensor; receiving GNSS measurements from the satellite via a GNSS receiver; detecting the NLOS reception from the GNSS measurements using the SWM; correcting the NLOS reception by NLOS reconstruction when no reflection point is found in the SWM; and estimating GNSS positioning using a least-squares algorithm.

In one embodiment, the method further includes minimizing the point cloud capacity of the SWM by excluding point clouds that are far from the GNSS receiver, such that the 3D point cloud is located within a sliding window.

In one embodiment, the step of generating the SWM includes: obtaining a local map from a LiDAR scan match based on the 3D point cloud from a 3D LiDAR sensor; and converting the SWM from the carrier coordinate system to the local northeast-sky (ENU) coordinate system using the orientation of the AHRS.

In one embodiment, the step of detecting the NLOS receiver from the GNSS measurements is performed using a fast search method, wherein the fast search method includes: initializing a search point at the center of the 3D LiDAR sensor; determining a search direction connecting the GNSS receiver and the satellite based on the satellite's elevation and azimuth angles; moving the search point along the search direction by a fixed increment _Δdpix ; calculating the number of neighboring points ( _Nk ) near the search point; and classifying the search point as an NLOS satellite if _Nk exceeds a predetermined threshold _Nthres .

In one embodiment, the method further includes: correcting the NLOS reception by re-estimating the GNSS measurements using a model calibration based on the SWM, wherein the SWM provides a dense, discrete, and unorganized 3D point cloud without continuous building surfaces or boundaries.

In one embodiment, model calibration includes: detecting the reflection point corresponding to the NLOS receiver using an efficient kdTree structure based on a reflector detection algorithm, wherein the reflector detection algorithm includes: traversing all azimuth angles from 0° to 360°, with an azimuth resolution of _αres and an elevation angle of [value missing], detecting potential reflectors when the line of sight between the connection point _pj and the satellite is not obstructed; and detecting the unique reflector with the shortest distance between the GNSS receiver and the potential reflector.

In one embodiment, NLOS uses a weighting scheme with a scaling factor to perform the NLOS reconstruction. The scaling factor is used to deweight the NLOS reception, and the weighting scheme includes the following definitions: if the satellite is classified as a LOS measurement, the scaling factor is calculated based on the satellite signal-to-noise ratio (SNR) and elevation angle; if the satellite is classified as an NLOS measurement and pseudorange error has been corrected, the scaling factor is calculated based on the satellite SNR and the elevation angle; and if the satellite is classified as an NLOS measurement but no reflection point is detected, the scaling factor is calculated based on the satellite SNR, the elevation angle, and the scaling factor _Kw .

This summary is provided to introduce the selected concepts in a simplified form, which will be further described in the detailed description below. This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter. Other aspects and advantages of the invention are disclosed as shown in the following embodiments.

Attached Figure Description

The accompanying drawings include figures for further illustration and elucidation of the above and other aspects, advantages, and features of this disclosure. It should be understood that these drawings depict only certain embodiments of the disclosure and are not intended to limit its scope. It should also be understood that these drawings are shown for simplicity and clarity and are not necessarily drawn to scale. This disclosure will now be described and explained with additional specificity and detail using the accompanying drawings, in which:

Figure 1 is a schematic diagram of a 3D LiDAR-assisted GNSS real-time dynamic differential positioning method according to an exemplary embodiment of the present disclosure;

Figure 2 is a schematic diagram of a 3D LiDAR-assisted GNSS NLOS mitigation method according to an exemplary embodiment of the present disclosure;

Figure 3 is an exemplary demonstration of a sliding window map and a real-time 3D point cloud generated according to an exemplary embodiment of the present disclosure;

Figure 4 is a schematic diagram of the coordinate system used in this invention.

Figure 5 is a conceptual diagram illustrating NLOS detection based on a generated sliding window map according to an exemplary embodiment of the present disclosure;

Figure 6 shows the reconstruction and correction of contaminated GNSS signals through the perception of the world;

Figure 7 shows the positioning performance before and after the application of the 3D LiDAR-assisted GNSS real-time dynamic differential positioning method according to an embodiment of the present invention.

Figure 8 is a system diagram of a 3D LiDAR-assisted GNSS NLOS mitigation method configured to be used in conjunction with a vehicle, according to an embodiment of the present disclosure.

Detailed Implementation

This disclosure is primarily directed to the field of autonomous driving or other types of autonomous systems requiring navigation using satellite positioning systems. The vehicle may be an ADV (Advanced Driver Assistance Vehicle) or a vehicle equipped with ADAS (Advanced Driver Assistance Systems). More specifically, but not limited to, this disclosure provides a 3D LiDAR-assisted GNSS NLOS mitigation method that can detect and correct NLOS signals in very deep urban canyons, and the system implements this method. The purpose of this disclosure is to provide a method for mitigating NLOS caused by static buildings and dynamic objects, thereby achieving high accuracy in highly dense urbanized areas.

Effects, advantages, solutions to problems, and any elements that may lead to or make more apparent any effects, advantages, or solutions should not be construed as key, essential, or fundamental features or elements of any or all claims. The invention is defined solely by the appended claims, including any modifications made during the pending period of this application and all equivalents of those claims.

In the following claims and the foregoing description of the invention, unless the context requires otherwise due to the language of expression or necessary implication, the word "comprise" (or variations thereof, such as "comprises" or "comprising") is used in an inclusive manner, that is, to specify the presence of the stated feature but not to exclude the presence or addition of further features in various embodiments of the invention.

As used herein, the term "Global Navigation Satellite System" or "GNSS" refers to a universal satellite-aided navigation system in which electronic receivers can determine their position with specified accuracy using line-of-sight radio signals transmitted by satellites. A GNSS receiver can receive and process signals from multiple satellites orbiting the Earth to determine the receiver's position and perform position transformations to determine the vehicle's position. For the purposes of this invention, unless otherwise stated, the status of the GNSS receiver and the positions of the satellites are represented in the ECEF coordinate system.

The term "LiDAR sensor" refers to a sensor that can measure the distance from a vehicle to an object by emitting a laser beam at the object and measuring the reflected portion of the laser beam to calculate the distance.

The term "Attitude Heading Reference System" or "AHRS" refers to a system with one or more sensors configured to use vibrations to measure changes in the vehicle's orientation, azimuth, and/or acceleration based on a vertical reference.

The first embodiment of this disclosure is to detect and exclude GNSS NLOS reception to further improve GNSS-RTK positioning. Simultaneously, using the improved GNSS-RTK positioning to correct 3D point cloud drift, thereby improving overall positioning accuracy, is also of great significance.

Figure 1 provides a schematic diagram of the 3D LiDAR-assisted GNSS-RTK positioning method proposed in this disclosure. The system comprises two parts: (1) generation of a real-time environment description based on the cloud from the 3D LiDAR and IMU, and correction from the GNSS-RTK solution S100A; and (2) GNSS NLOS detection and elimination based on the real-time environment description, and GNSS-RTK positioning based on the surviving S100B satellites.

In some embodiments, the use of a 3D LiDAR sensor 110 improves GNSS-RTK positioning in urban canyons by fundamentally addressing the problems of conventional GNSS-RTK caused by signal reflection and occlusion. First, the 3D LiDAR sensor 110 and LIO 120 are executed, receiving and loosely integrating LiDAR factors and IMU factors using local factor map optimization 141 to estimate the relative motion between two epochs and generate a 3D PCM to provide an environmental description. The LiDAR factors are obtained from LiDAR scan matching 111, and the IMU factors are obtained from pre-integration 121. The environmental description used here is a local environmental description. In some embodiments, the 3D LiDAR sensor 110 is a Velodyne 32 configured to collect raw 3D point cloud data at a frequency of 10 Hz, and the LIO 120 is an Xsens Ti-10 IMU configured to collect data at a frequency of 100 Hz. Simultaneously, the positions of the surrounding point cloud are estimated. Therefore, PCM correction 142 is performed using the position estimation of the surrounding point cloud on the 3D PCM. The effect is that this method can advantageously generate locally accurate PCMs.

Secondly, potential GNSS NLOS satellites are detected and excluded with the aid of environmental description. GNSS receiver 130 receives GNSS measurements from the satellites, which can be used to derive an initial guess of the floating-point solution using a conventional least-squares algorithm 131. In some embodiments, GNSS receiver 130 is a commercial-grade u-blox F9P GNSS receiver used to collect raw GPS/BeiDou measurements at a data frequency of 1 Hz. Furthermore, a NovAtel SPAN-CPT, GNSS (GPS, GLONASS, and BeiDou) RTK/INS (fiber optic gyroscope, FOG) integrated navigation system can be used to provide ground-based positioning. All data is collected and synchronized using ROS.

For subsequent GNSS NLOS detection, the first embodiment of this disclosure does not rely on an initial guess of the floating-point solution of the GNSS receiver 130. Instead, GNSS NLOS reception is detected from GNSS measurements using 3D PCM generated by LiDAR or inertial integration and correlation algorithms. Therefore, using GNSS NLOS exclusion 151 can alleviate the problem of poor GNSS measurement quality, which essentially excludes potential GNSS NLOS reception to obtain surviving GNSS satellite measurements.

To address this issue, this disclosure proposes using landmarks from the generated point cloud map as "auxiliary landmark satellites" to fundamentally improve satellite geometry. Advantageously, the auxiliary landmark satellite and the receiving satellite complement each other as point cloud maps, providing low-elevation auxiliary landmark satellites. Such low-elevation auxiliary landmark satellites are typically unavailable for physically receiving satellites. Therefore, contaminated GNSS NLOS satellites are excluded, and satellite geometry is improved with the help of auxiliary landmark satellites.

Next, a floating-point solution can be obtained by performing GNSS-RTK floating-point estimation 152, followed by ambiguity resolution 153 to obtain a fixed ambiguity solution, which is estimated based on surviving GNSS satellite measurements. In some embodiments, ambiguity resolution 153 is performed by applying the LAMBDA algorithm. Finally, the estimated fixed GNSS-RTK positioning solution 154 is fed back to the 3D LiDAR sensor and LIO to perform PCM correction 142 to further correct for 3D point cloud drift. By combining improved GNSS-RTK positioning assisted by GNSS NLOS detection, this method can essentially correct the drift of 3D point cloud maps generated by LiDAR or inertial integration. Therefore, the proposed method effectively combines the complementarity of LIO (which provides environmental description for GNSS NLOS detection in a locally accurate manner over a short time) and GNSS-RTK (drift-free, with global reference positioning but affected by GNSS NLOS).

The first embodiment of this disclosure addresses this problem by utilizing the sensing capabilities of a 3D LiDAR sensor to detect and exclude potential GNSS NLOS reception. This method achieves an accuracy of 10 centimeters even in urban canyon scenarios, meeting the navigation requirements of autonomous driving. In the second embodiment of this disclosure, an alternative method and system are provided that employ 3D LiDAR to assist in GNSS point positioning. Positioning accuracy is limited to a range of only 5 meters, which is less accurate than the GNSS-RTK of the first embodiment.

Referring to Figure 2, a schematic diagram of a 3D LiDAR-assisted GNSS NLOS mitigation method is illustrated. The system consists of two parts: (1) a real-time SWM generation S200A based on 3D point clouds from 3D LiDAR sensors and AHRS; and (2) a GNSS NLOS detection and correction S200B based on real-time environment description.

Advantageously, the SWM is generated based on real-time 3D point clouds from the 3D LiDAR sensor 110, where only the 3D point cloud within a sliding window is used to generate the SWM, since point clouds far from the GNSS receiver 130 are unnecessary for NLOS detection. Therefore, the size of the SWM can be minimized. The SWM is first generated using the 3D LiDAR sensor 110 and the AHRS 160, where the SWM essentially provides an environmental description for detecting and correcting NLOS reception. A local map is obtained from LiDAR scan matching 211 based on the 3D point cloud from the 3D LiDAR sensor 110. Orientation from the AHRS 160 is directly adopted to transform the SWM from the carrier coordinate system to the local ENU coordinate system.

By performing local transformation 241, a SWM can be generated, which is executed in real time for GNSS NLOS detection. This is crucial for achieving better environmental description capabilities, thereby enhancing the FOV of LiDAR sensing. In conventional methods, only real-time 3D point clouds are applied to further detect NLS satellites. However, the NLOS detection capability is limited by the FOV of the 3D LiDAR sensor 110. In this regard, the second embodiment of this disclosure improves upon the shortcomings of the prior art by accumulating real-time point clouds into the SWM. The FOV of the 3D LiDAR sensor 110 can be effectively enhanced.

Referring to Figure 3, which illustrates the perceived digital world, the differences between real-time 3D LiDAR point clouds and SWM (Surveyed Width Model) are shown. Perceiving the environment opens a new window to reveal the effects of GNSS signal transmission obstruction or reflection caused by surrounding tall buildings, etc. The 3D LiDAR sensor 110 has a limited FOV (Field of View), and a single frame of 3D point cloud can only effectively detect low-lying areas of buildings and vehicles (e.g., double-decker buses). Simply relying on real-time 3D point clouds cannot effectively classify the visibility of high-elevation satellites. Due to the physical scanning angle distribution of the 3D LiDAR sensor 110, the real-time 3D point cloud is also sparse.

The aforementioned problems can be effectively improved using the SWM disclosed herein. The map points of the SWM are shown in Figure 3 (ground points are removed from the SWM for effective NLOS detection). With the help of the SWM, the cutoff elevation angle can reach 76°, thus enabling the classification of visibility for satellites with elevation angles less than 76°. The point cloud in the SWM is much denser than the original real-time 3D point cloud, which significantly improves the accuracy of NLOS detection. A snapshot of the complete SWM map is shown in the upper right corner of Figure 3. Buildings and dynamic objects, such as double-decker buses and even trees, are included in the SWM, which are not included in the 3D building model.

To generate point cloud maps based on real-time 3D point clouds, the SLAM method is typically used. In fact, satisfactory accuracy with low drift can be achieved in a short time. However, errors accumulate over time, leading to significant errors after long-term travel, and loop closure is usually impossible. Therefore, in practical applications, only objects within a circle with a radius of approximately 250 meters will be received by GNSS NLOS, while distant buildings are ignored.

Referring to Figure 4, the coordinate system transformation is illustrated. The ECEF coordinate system is fixed at the center of the Earth. The first point is chosen as the reference for the ENU coordinate system, while the extrinsic parameters between the LiDAR, AHRS, and GNSS receivers are pre-fixed and calibrated. The L coordinate system (ENU) is transformed into the G coordinate system (ECEF).

This disclosure uses only the last Nsw coordinate system of the 3D point cloud to generate the sliding window map. As a result, the drift error in map generation is thus limited to a very small value and is determined by the window size (Nsw). To minimize significant vertical drift in urban canyons with a large number of dynamic objects, this disclosure employs absolute ground to constrain vertical drift. Details of the SWM generation algorithm are demonstrated below:

Input: A series of point clouds from epoch tN _sw +1 to epoch t, representing external parameters between 3D LiDAR, AHRS, and GNSS receivers.

Output:

Step 1: Initialization

Step 2: SWM generation:

First, align all point clouds to a local map using a given coordinate system as the primary coordinate system. Second, use the following equation to transform a point within the local map to the receiver carrier's coordinate system.

Third, use the following equation to convert a point within a local map to the ENU coordinate system.

Returning to Figure 2, the second part of the GNSS pseudorange measurement NLOS detection and correction S200B setup has four stages. First, GNSS NLOS detection 251 is used for model validation based on satellite visibility classification using SWM, which effectively identifies LOS and NLOS satellites. Next, if a satellite is classified as NLOS, GNSS NLOS correction 252 is performed as a model calibration stage, which re-estimates the GNSS measurements by correcting the NLOS pseudorange measurements (CNLOS). However, if a satellite is classified as NLOS but its reflector is not within SWM (FNLOS), NLOS correction is unavailable. The third stage is NLOS reconstruction 253, a model repair stage, which deweights the NLOS measurements for further positioning. GNSS positioning is estimated based on the pseudorange measurements using a least squares algorithm 254.

GNSS NLOS detection performed through model validation 251

No object detection algorithm is needed to recover the actual height of detected dynamic objects or building surfaces. Since SWM only provides unorganized discrete points, a fast search method based on real-time SWM can be used to directly detect NLOS reception, eliminating the need for an object detection process.

The inputs to the fast search method include: the maximum search distance D _{_thres} at the azimuth epoch of the elevation angle of satellite s and a constant increment value Δd_{_pix}. The fast search method is used to determine the satellite visibility of satellite s. In step 1, the search point is initialized in the ENU coordinate system by the 3D LiDAR center shown. The search direction connecting the GNSS receiver 130 and the satellite is determined by the satellite's elevation and azimuth angles. The SWM is converted into a kdTree structure for finding neighboring points. A kdTree is a special structure for point cloud processing that can be performed efficiently when searching for neighboring points. In step 2, given a fixed increment value Δd_{_pix} , the search point is moved to the next point according to the search direction exemplarily shown in Figure 5, using the following equation.

k represents the index of the search point. The number of neighboring points ( _Nk ) near the search point is calculated. According to step 3, if _Nk exceeds a predetermined threshold _Nthres , then there are map points near the search point originating from buildings or moving objects, and the line-of-sight between the GNSS receiver 130 and the satellite is considered blocked. Therefore, the satellite s is classified as an NLOS satellite. Otherwise, steps 2 and 3 are repeated. If _kΔdpix > _Dthres , then the direction between the GNSS receiver 130 and the satellite is along the LOS. _Dthres can be set to define a distance where locations within that distance are considered for NLOS detection. Satellite visibility can be classified using a fast search method.

A demonstration of satellite visibility classification results with NLOS and LOS satellites is shown in Figure 6. Gray circles represent contaminated (multipath effect) GNSS satellites, and white circles represent healthy GNSS satellites. The number provided next to each circle indicates the elevation angle of each satellite. In this example, an elevation angle of 54° was detected. As previously mentioned, the maximum cutoff elevation angle is 76°, and SWM-based elevation angles can be significantly correlated with street width. Narrower streets can achieve higher cutoff elevation angles. However, low elevation angles of NLOS satellites can cause most GNSS positioning errors. This disclosure is provided to detect and recover signal obstruction and reflections for highly accurate GNSS positioning, even in challenging urban canyons.

GNSS NLOS correction via model calibration 252

NLOS correction is performed through SWM-based model calibration. To effectively estimate potential NLOS errors, the distance between the GNSS receiver, satellite elevation, and azimuth needs to be based on an NLOS error model. Specifically, model calibration involves detecting reflection points corresponding to the NLOS receiver. Traditionally, ray tracing techniques are used to simulate NLOS signal transmission paths to locate NLOS reflectors. However, this technique suffers from high computational costs. This disclosure provides a method that does not produce continuous building surfaces and sharp building boundaries. Instead, SWM only provides a large amount of dense, discrete, and unorganized point clouds. Reflector search from the SWM is performed using an efficient kdTree structure based on a reflector detection algorithm.

The inputs to the reflector detection algorithm used to perform model calibration include: the azimuth angle at elevation epoch t of satellite s and the azimuth resolution _αres . The output is the nearest reflection point, which is the most likely reflecting surface of NLOS satellites s.

In step 1, the search point is initialized at the center of the 3D LiDAR. The search direction is determined based on the satellite elevation and azimuth angles.

A typical signal reflection path consists of two segments. The first segment is the signal transmission from the satellite to the reflector. The second segment is the signal transmission from the reflector to the GNSS receiver 130. Since the reflected signal should have the same elevation angle as the expected directional signal, step 2 of the reflector detection algorithm involves traversing all azimuth angles from 0° to 360°, with an azimuth resolution of α_{_res} and an elevation angle of α_el, to find all possible NLOS transmission paths.

In step 3, if the line-of-sight associated with the direction is blocked by point _pj , then point _pj is likely a reflecting point. Specifically, if the line-of-sight connecting point _pj and the satellite is not blocked, then point _pj is considered a possible reflector and saved.

In step 4, steps 2 and 3 are repeated until _αs > 360°, thereby identifying all possible reflectors based on the assumption of the same elevation angle.

Finally, step 5, based on the shortest distance assumption, detects the unique reflector from the shortest distance between the GNSS receiver 130 and the reflector.

Advantageously, the reflector detection algorithm does not depend on the detection accuracy of the building surface. The short-range assumption effectively prevents overcorrection because only the nearest reflector is identified as the unique reflector. The potential NLOS delay of the satellites can be calculated as:

The operator ||*|| is used to calculate the norm of a given vector. sec(*) denotes the secant function. Due to the sparsity of SWM, although it is still denser than 3D real-time point clouds, there are still some satellites whose reflectors cannot be found using SWM. Therefore, if a satellite is classified as NLOS but its reflector cannot be found within SWM, NLOS reconstruction is recommended.

NLOS Reconstruction for Model Repair 253

Satellites with low elevation angles and low SNRs are more likely to be contaminated by NLOS errors. Traditional pseudorange uncertainty modeling methods based on satellite elevation angle and SNR can produce satisfactory performance in open areas, but not in deep urban canyons. In particular, weighted schemes treat LOS and NLOS in the same way, which is undesirable when NLOS has been detected. This disclosure corrects GNSS NLOS reception by performing NLOS reconstruction, which essentially models the uncertainties of LOS and NLOS using a weighted scheme in which a scaling factor is added to deweight the NLOS measurements. The weighted scheme includes: (1) calculating a scaling factor based on satellite SNR and elevation angle if the satellite is classified as a LOS measurement; (2) calculating a scaling factor based on satellite SNR and elevation angle if the satellite is classified as an NLOS measurement and pseudorange errors have been corrected; and (3) calculating a scaling factor based on satellite SNR and elevation angle and a scaling factor _Kw if the satellite is classified as an NLOS measurement but no reflection point has been detected.

GNSS positioning using the least squares algorithm 254

The pseudorange measurement from the GNSS receiver is expressed as:

in:

It is the geometric range between the satellite and the GNSS receiver;

It is the ionospheric delay distance;

It is the tropospheric delay distance; and

It is noise caused by multipath effect, NLOS reception, receiver noise and antenna phase-dependent noise.

Simultaneously, the traditional Saastamoinen model and the Klobuchar model are used to compensate for atmospheric effects (and), respectively. The observation model for GNSS pseudorange measurements from a given satellite s is expressed as:

Among them, there is the associated noise.

Suppose that the Jacobian matrix, after subtracting the NLOS error observation function from it before further GNSS positioning, can be expressed as:

Where m represents the total number of satellites in epoch t.

The location of the GNSS receiver can be estimated iteratively using weighted least squares as follows:

Where W_{_t} represents the weight matrix based on the weights estimated in NLOS Reconstruction 253, as shown below:

in,

The weights defined for calculating LOS measurements based on satellite SNR and elevation angle are as follows:

Where T represents the signal-to-noise ratio threshold; a, A, and F are predetermined.

Figure 7 shows the positioning performance before and after applying the 3D LiDAR-assisted GNSS real-time dynamic differential positioning method developed in the first embodiment of the present invention. Traditional GNSS solutions have an error of approximately 30 meters, leading to lane misjudgments when vehicles are driving in urban canyons. This improved technology can provide powerful and accurate positioning results to support the operation of L4 autonomous vehicles.

Experimental results

To verify the effectiveness of the proposed method, two experiments were conducted in typical urban canyons with static buildings, trees, and dynamic objects (such as double-decker buses). The first experiment was conducted in urban canyon 1, with a street width of 22 meters and a building height of 35 meters. The second experiment was conducted in urban canyon 2, with a street width of 12.1 meters and a building height of 65 meters.

The vehicle is equipped with a u-blox M8TGNSS receiver for collecting raw GNSS measurements at a frequency of 1 Hz. The 3D LiDAR sensor is a Velodyne 32, configured to collect raw 3D point clouds at a frequency of 10 Hz. An Xsens Ti-10 INS is used to collect data at a frequency of 100 Hz. Additionally, a NovAtel SPAN-CPT, a GNSS (GPS, GLONASS, and BeiDou) RTK/INS (fiber optic gyroscope, FOG) integrated navigation system, is used to provide ground-based positioning. The FOG's gyro zero-bias operation stability is 1 degree per hour, and its random walk is 0.067 degrees per hour. The baseline between the rover and the GNSS base station is approximately 7 kilometers. All data is collected and synchronized using ROS. The following five configurations were evaluated: (1) u-blox GNSS receiver only; (2) weighted least squares (WLS); (3) weighted least squares excluding all NLOS satellites (WLS-NE); (4) weighted least squares with reweighting of all NLOS satellites (R-WLS); and (5) weighted least squares with NLOS correction if a reflector is detected, and reweighting of NLOS satellites if no reflector is detected (CR-WLS).

Table 1: Localization performance of GNSSSPP in City Canyon 1

The results of GNSS positioning experiments using five methods are shown in Table 1 above. The average positioning error using the u-blox receiver was 31.02 meters, with a standard deviation of 37.69 meters. The maximum error reached 177.59 meters due to severe multipath interference from surrounding buildings and NLOS reception. For WLS, the positioning error decreased to 9.57 meters, with a standard deviation of 7.32 meters. The maximum error also decreased to less than 50 meters. After excluding all detected non-line-of-sight satellites, the positioning error increased to 11.63 meters, worse than WLS. This occurred because excessive NLOS exclusion significantly distorts the perceived geometry of the satellites. Availability decreased slightly from 100% to 96.01%. Therefore…

Table 2: Location performance of GNSSSPP in City Canyon 2

In the more challenging scenario of Urban Canyon 2, some NLOS satellites reflected by buildings taller than 40 meters were found to be undetectable. A positioning error of 30.68 meters was obtained using a u-blox receiver, with a maximum error of 92.32 meters. Based on raw pseudorange measurements from the u-blox receiver, a GNSS positioning error of 23.79 meters was obtained using WLS. Compared to GNSS positioning using data directly from the u-blox receiver, the maximum error increased slightly to 104.83 meters. After excluding all detected NLOS satellites from GNSS positioning (WLS-NE), the mean and standard deviation increased to 25.14 meters and 23.73 meters, respectively. Due to the lack of satellites for GNSS positioning, the availability of GNSS positioning data dropped to 95.52%, again demonstrating that completely excluding NLOS in Urban Canyon is not advisable. With the aid of NLOS reconstruction, the 2D error using R-WLS was reduced to 19.61 meters, guaranteeing 100% availability. The GNSS positioning error was further reduced to 17.09 meters using the CR-WLS method. The improved results demonstrate the effectiveness of the proposed 3D LiDAR-assisted GNSS positioning method. The maximum error still reaches 71.28 meters because not all NLOS satellites can be detected and mitigated.

Figure 8 is a system diagram of a possible implementation and combination of a 3D LiDAR-assisted GNSS NLOS mitigation method in a vehicle according to an exemplary embodiment of the present disclosure. The vehicle uses LiDAR-assisted GNSS to support vehicle positioning using a satellite positioning system. In some embodiments, the system includes a 3D LiDAR sensor 110, an AHRS 160, a GNSS receiver 130, a processor 810, a memory 820, a user interface 830, autonomous control 840, and a communication interface 850. In some embodiments, without departing from the scope and spirit of the present disclosure, the 3D LiDAR sensor 110, the AHRS 160, and/or the GNSS receiver 130 may be integrated into the processor 810. In this case, the processor 810 is a dedicated device with built-in components capable of receiving GNSS satellite signals, direction, and/or local 3D point cloud maps.

Processor 810 may be one or more general-purpose processors, special-purpose processors, digital signal processing chips, application-specific integrated circuits (ASICs), or other processing structures configured to perform one or more of the methods described above. Processor 810 is communicatively connected to 3D LiDAR sensor 110, AHRS 160, and GNSS receiver 130 for receiving local maps, orientation, and GNSS measurements, respectively. Processor 810 may include, or communicate with, memory 820 (as a discrete component), for storing positioning data and/or other received signals and retrieving previous keyframe point cloud data of the SWM that enhances the FOV of 3D LiDAR sensor 110.

In some embodiments, memory 820 may include, but is not limited to, solid-state storage devices, such as random access memory (RAM) and/or read-only memory (ROM), the ROM being programmable, flash-updatable, and/or the like. Such storage devices can be configured to implement any suitable data storage, including but not limited to various file systems, database structures, etc. Memory 820 may also include software instructions configured to cause processor 810 to perform one or more functions according to the methods of this disclosure. Thus, the functions and methods described above can be implemented as computer code and/or instructions executable by processor 810. Then, in one aspect, such code and/or instructions can be used to configure and/or adapt a computer (or other computer device) to perform one or more operations according to the described methods. Memory 820 may therefore include a non-transitory machine-readable medium having instructions and/or computer code embedded therein. Common forms of computer-readable media include, for example, hard disks, magnetic or optical media, RAM, PROM, EPROM, FLASH-EPROM, USB memory sticks, any other memory chip or cartridge, carrier waves, or any other medium from which a computer can read instructions and/or code.

User interface 830 allows interaction with the driver, passengers, and/or other individuals controlling the vehicle, and may include one or more input devices selected from the group consisting of a touchscreen, touchpad, buttons, switches, microphone, etc. The driver or person controlling the vehicle can use user interface 830 to activate or deactivate autonomous control 840, or select different operating modes and/or the degree of autonomous driving.

The autonomous control unit 840 is an output device for controlling vehicles based on GNSS positioning. The processor 810 can thus determine the position of vehicles, static buildings, and dynamic objects, enabling the vehicle to adjust its navigation route, speed, acceleration, onboard alarm system, and/or other functions accordingly. The autonomous control unit 840 can therefore implement various functions of intelligent transportation systems, such as autonomous driving, semi-autonomous driving, and navigation.

Communication interface 850 provides an interface that allows data to communicate with networks, vehicles, positioning servers, servers, wireless access points, other computer systems, and/or any other electronic devices described herein. In some embodiments, communication interface 850 may include, but is not limited to, modems, network interface cards, infrared communication devices, wireless communication devices, and/or chipsets (e.g., Bluetooth devices, IEEE 802.11 devices, IEEE 802.15.4 devices, Wi-Fi devices, WiMAX devices, cellular communication facilities, etc.).

This system diagram is intended to provide a general illustration of 3D LiDAR-assisted GNSS, while other components and system blocks may be included as needed. One or more components within the system may be integrated or partitioned and may be located in different physical locations within the vehicle, or otherwise remotely deployed on a server or network system. Override control (not shown) or other safety mechanisms may also be provided to allow the driver or passengers of the vehicle to take over control of the vehicle in an emergency.

As demonstrated, the method of this disclosure provides an effective way to eliminate potential GNSS NLOS reception using either an environment description generated by a 3D LiDAR sensor and LIO, or a SWM using a 3D LiDAR sensor and AHRS. This method achieves accurate positioning even in urban canyons for autonomous driving. This invention addresses the problem of insufficient positioning accuracy hindering the deployment of autonomous driving applications. Above all these existing positioning solutions, GNSS-RTK is an indispensable method capable of providing global reference positioning in sparse scenes. However, its accuracy cannot be guaranteed in urban canyons due to signal obstruction and reflections leading to GNSS NLOS reception. This disclosure solves this problem by leveraging the sensing capabilities of 3D LiDAR to detect and eliminate potential GNSS NLOS reception. In some embodiments, the method provided herein achieves higher accuracy even in urban canyon scenarios to meet the navigation requirements of autonomous driving.

The above method can be directly applied to the autonomous driving industry. Specifically, even in urban canyons, the present invention can be used to provide accurate positioning solutions. Furthermore, the method can also be applied to other autonomous systems with navigation needs, such as mobile robots and drones. Simultaneously, the method can also be used for geodetic mapping of urban canyons. This illustrates a 3D LiDAR-assisted global navigation satellite system with improved positioning performance according to this disclosure. Obviously, variations and other features and functions or alternatives thereof disclosed above can be combined into many other different configurations, devices, apparatuses, and systems. Therefore, this embodiment is to be considered illustrative rather than restrictive in all respects. The scope of this disclosure is defined by the appended claims rather than by the foregoing description and is intended to include all changes falling within the equivalent meaning and scope of the claims.

List of abbreviations

3D

3DMA 3D Map Building Assist

ADAS (Advanced Driver Assistance System)

ADV (Autonomous Vehicle)

AHRS Attitude and Heading Reference System

DOP Precision Attenuation Factor

ECEF Geocentric-Earth-Fixed Coordinate System

ENU Northeast

FOV (Field of View)

GNSS Global Navigation Satellite System

GNSS-RTK Global Navigation Satellite System Real-time Dynamic Differential Positioning

GPS Global Positioning System

LiDAR (Light Detection and Ranging)

LIO LiDAR - Inertial Odometry

IMU (Inertial Measurement Unit)

LOS (Location of View)

NLOS (Non-line-of-sight)

PCM point cloud map

ROS Robot Operating System

RTK Real-time Dynamic Differential Positioning

SLAM (Simultaneous Localization and Mapping)

SWM sliding window map

t GNSS epoch

GECEF coordinate system

LENU coordinate system

s Satellite Index

r GNSS receiver

BI AHRS Carrier Coordinate System

BL LiDAR carrier coordinate system

BR GNSS receiver carrier coordinate system

ρ pseudo-distance

pseudorange of the satellite at epoch t

The position of the satellite at epoch t

Location of satellite GNSS receiver at epoch t

_δr,t GNSS receiver clock bias

Satellite clock bias

Signal-to-noise ratio (SNR)

_Kw is the weighting scaling factor.

Satellite elevation angle

Satellite azimuth

k Search point index

N _k neighboring points

Claims (11)

1. A method for a vehicle to position itself using a satellite positioning system, the method comprising: A sliding window map (SWM) is generated in real time based on 3D point clouds from a 3D LiDAR sensor and an attitude heading reference system (AHRS), wherein the SWM provides an environmental description for detecting and correcting non-line-of-sight (NLOS) reception. The point cloud capacity of the SWM is minimized by excluding point clouds that are far from the GNSS receiver, so that the 3D point cloud is located within the sliding window. The 3D point cloud from the previous frame is accumulated into the SWM to enhance the field of view (FOV) of the 3D LiDAR sensor; Receive Global Navigation Satellite System (GNSS) measurements from satellites via a GNSS receiver; The NLOS reception is detected from the GNSS measurements using the SWM; When no reflection point is found in the SWM, the NLOS reception is corrected by NLOS reconstruction; and GNSS positioning is estimated using the least squares algorithm; The NLOS reconstruction is performed using a weighting scheme with a scaling factor, wherein the scaling factor is used to deweight the NLOS reception, and the weighting scheme includes the following definitions: If the satellite is classified as a LOS measurement, the scaling factor is calculated based on the satellite signal-to-noise ratio (SNR) and elevation angle. If the satellite is classified as an NLOS measurement and pseudorange error is corrected, then the scaling factor is calculated based on the satellite's SNR and elevation angle; and If the satellite is classified as an NLOS measurement but no reflection point is detected, the scale factor is calculated based on the satellite SNR, the elevation angle, and the scale factor. 2. The method of claim 1, wherein the step of generating the SWM comprises: Based on the 3D point cloud from the 3D LiDAR sensor, a local map is obtained from LiDAR scan matching; and The SWM is transformed from the carrier coordinate system to the local northeast-sky ENU coordinate system using the AHRS orientation. 3. The method of claim 1, wherein the step of detecting the NLOS reception from the GNSS measurements is performed using a fast search method, wherein the fast search method comprises: Initialize the search point at the center of the 3D LiDAR sensor; The search direction connecting the GNSS receiver and the satellite is determined based on the satellite's elevation and azimuth angles; The search point is moved along the search direction by a fixed increment. Calculate the number of neighboring points near the search point; and If the search point exceeds a predetermined threshold, it will be classified as an NLOS satellite. 4. The method of claim 1, further comprising: correcting the NLOS reception by re-estimating the GNSS measurements using a model calibration based on the SWM, wherein the SWM provides a dense, discrete, and unorganized 3D point cloud without continuous building surfaces or boundaries. 5. The method according to claim 4, wherein the model calibration comprises: The reflection point corresponding to the NLOS receiver is detected using an efficient kdTree structure based on a reflector detection algorithm, wherein the reflector detection algorithm includes: Iterate through all azimuth angles from 0 ^° to 360 ^° , with azimuth resolution of and elevation angle of . Detect potential reflectors when the line of sight between the connection point and the satellite is unobstructed; and The only reflector with the shortest distance between the GNSS receiver and the potential reflector is detected. 6. A LiDAR-assisted global navigation satellite system for use by a vehicle, used to support the vehicle's positioning using a satellite positioning system, the system comprising: 3D LiDAR sensor; Attitude and Heading Reference System (AHRS); A Global Navigation Satellite System (GNSS) receiver, configured to receive GNSS measurements from satellites; A processor, communicatively connected to the 3D LiDAR sensor, the AHRS, and the GNSS receiver, wherein the processor is configured to: A sliding window map (SWM) is generated in real time based on 3D point clouds from a 3D LiDAR sensor and the AHRS, wherein the SWM provides an environmental description for detecting and correcting non-line-of-sight (NLOS) reception. The point cloud capacity of the SWM is minimized by excluding point clouds that are far from the GNSS receiver, so that the 3D point cloud is located within the sliding window. The 3D point cloud from the previous frame is accumulated into the SWM to enhance the field of view (FOV) of the 3D LiDAR sensor; The SWM is used to detect NLOS reception from the GNSS measurements of the GNSS receiver; When no reflection point is found in the SWM, the NLOS reception is corrected by NLOS reconstruction; and GNSS positioning is estimated using the least squares algorithm; The NLOS reconstruction is performed using a weighting scheme with a scaling factor, wherein the scaling factor is used to deweight the NLOS reception, and the weighting scheme includes the following definitions: If the satellite is classified as a LOS measurement, the scaling factor is calculated based on the satellite signal-to-noise ratio (SNR) and elevation angle. If the satellite is classified as an NLOS measurement and pseudorange error is corrected, then the scaling factor is calculated based on the satellite's SNR and elevation angle; and If the satellite is classified as an NLOS measurement but no reflection point is detected, the scale factor is calculated based on the satellite SNR, the elevation angle, and the scale factor. 7. The system of claim 6, wherein the processor is configured to obtain a local map based on a 3D point cloud from the 3D LiDAR sensor; and to convert the SWM from the carrier coordinate system to the local northeast-sky ENU coordinate system using the orientation of the AHRS. 8. The system of claim 6, wherein the processor is configured to detect the non-line-of-sight reception from the GNSS measurements using a fast search method, wherein the fast search method comprises: Initialize the search point at the center of the 3D LiDAR sensor; The search direction connecting the GNSS receiver and the satellite is determined based on the satellite's elevation and azimuth angles; The search point is moved along the search direction by a fixed increment. Calculate the number of neighboring points near the search point; and If the search point exceeds a predetermined threshold, it will be classified as an NLOS satellite. 9. The system of claim 6, wherein the processor is further configured to correct the NLOS reception by re-estimating the GNSS measurements using a model calibration based on the SWM, wherein the SWM provides a dense, discrete, and unorganized 3D point cloud without continuous building surfaces or boundaries. 10. The system of claim 9, wherein the model calibration comprises: The reflection point corresponding to the NLOS receiver is detected using an efficient kdTree structure based on a reflector detection algorithm, wherein the reflector detection algorithm includes: Iterate through all azimuth angles from 0 ^° to 360 ^° , with azimuth resolution of and elevation angle of . Detect potential reflectors when the line of sight between the connection point and the satellite is unobstructed; and The only reflector with the shortest distance between the GNSS receiver and the potential reflector is detected. 11. The system of claim 6, further comprising: Vehicle autonomous control based on GNSS positioning enables various functions of the intelligent transportation system. The user interface for activating or deactivating the autonomous control; and Communication interface.