Radar point cloud error evaluation method, device, medium and product

Abstract

This application relates to the field of electronic technology, specifically to a radar point cloud error assessment method, device, medium, and product. The method includes: acquiring target tag data, vehicle RTK data, target vehicle RTK data, and a point cloud result file; obtaining the true values ​​of the motion parameters of the target vehicle relative to the vehicle based on the vehicle RTK data and the target vehicle RTK data; when the longitudinal distance between the center of the detection frame and the vehicle is less than or equal to a first distance threshold, based on the point cloud data of reflection points within the detection frame, determining reflection points within the detection frame whose absolute azimuth angle relative to the vehicle satisfies a first angle condition as candidate points, and determining candidate points whose point cloud intensity satisfies a first intensity condition as target points, with the longitudinal direction being the vehicle's driving direction; comparing the estimated values ​​of the motion parameters of the target points in the point cloud data with the true values ​​of the motion parameters to determine the point cloud error. Through the embodiments of this application, the accuracy and reliability of point cloud error assessment can be improved.

Description

Technical Field

[0001] This application relates to the field of electronic technology, and in particular to a radar point cloud error assessment method, device, medium, and product. Background Technology

[0002] Error assessment of point cloud data from vehicle-mounted millimeter-wave radar typically requires comparing the target motion parameters output by the radar with high-precision true values.

[0003] Currently, a direct approach is to take the mean or median of the longitudinal velocities of multiple reflection points within the target detection box as an estimate of the target vehicle's longitudinal velocity, and then compare it with the true value to assess point cloud error. However, in close-range scenarios, the target vehicle's reflection points are densely packed and affected by factors such as wheel rotation and ground clutter, resulting in significant differences in the longitudinal velocities of each reflection point within the detection box. Using the mean or median value for statistical analysis can easily deviate from the target vehicle's true velocity, leading to inaccurate error assessment results. Summary of the Invention

[0004] To address the aforementioned technical problems, embodiments of this application provide a radar point cloud error assessment method, device, medium, and product, which can improve the accuracy and reliability of point cloud error assessment.

[0005] In a first aspect, this application provides a radar point cloud error assessment method, comprising: acquiring target label data, vehicle RTK data, target vehicle RTK data, and a point cloud result file, wherein the target label data includes information of the target vehicle's detection frame and a first timestamp; the vehicle RTK data and the target vehicle RTK data are respectively data corresponding to the first timestamp obtained by nearest neighbor time matching based on the first timestamp; the point cloud result file includes point cloud data of multiple reflection points corresponding to the first timestamp obtained by time matching based on the first timestamp; obtaining the true values ​​of motion parameters of the target vehicle relative to the vehicle based on the vehicle RTK data and the target vehicle RTK data; when the longitudinal distance between the center of the detection frame and the vehicle is less than or equal to a first distance threshold, based on the point cloud data of the reflection points within the detection frame, determining the reflection points within the detection frame whose absolute azimuth angle relative to the vehicle satisfies a first angle condition as candidate points, and determining the candidate points whose point cloud intensity satisfies a first intensity condition as target points, wherein the longitudinal direction is the driving direction of the vehicle; comparing the estimated values ​​of motion parameters of the target points in the point cloud data with the true values ​​of motion parameters to determine the point cloud error.

[0006] In this embodiment, in close-range scenarios, reflection points within the detection box whose absolute azimuth angles meet the criteria are selected based on angle conditions. Reflection points near the center region of the target vehicle's side facing the vehicle are retained, avoiding interference from edge reflection points on velocity estimation. Then, reflection points whose point cloud intensity meets the criteria are selected as target points based on intensity conditions, ensuring the reliability of velocity estimation. The angle and intensity conditions work together to filter reflection points sequentially, ultimately using the velocity of a single target point to represent the target vehicle's velocity. This overcomes the problem of accurately estimating target velocity in close-range scenarios due to the dense reflection points and discrete velocities, improving the accuracy and reliability of point cloud error assessment.

[0007] In one possible implementation of the first aspect, the first angle condition is: the absolute value of the azimuth angle is the N smallest absolute values ​​of the azimuth angle among all reflection points within the detection box, where N is a positive integer.

[0008] In one possible implementation of the first aspect, there are multiple candidate points, and the first intensity condition is: the point cloud intensity is the largest among the multiple candidate points.

[0009] In one possible implementation of the first aspect, the method further includes: when the longitudinal distance between the center of the detection box and the vehicle is greater than a first distance threshold, performing quality screening and outlier removal on the reflection points within the detection box to determine the target point set; comparing the average estimated value of the motion parameters of the target point set in the point cloud data with the true value of the motion parameters to determine the point cloud error. Thus, in long-distance scenarios, by removing unreliable reflection points through quality screening and removing reflection points with abnormal speeds through outlier removal, the interference of sparse noise and outliers in long-distance scenarios is effectively suppressed, enhancing the stability of point cloud velocity estimation.

[0010] In one possible implementation of the first aspect, quality screening and outlier removal are performed on the reflection points within the detection frame to determine the target point set. This includes: screening the reflection points within the detection frame based on the signal stability and confidence level of each reflection point belonging to a moving target to obtain a candidate reflection point set; identifying outlier reflection points whose radial velocities do not meet the first velocity condition based on the radial velocities of each reflection point in the candidate reflection point set; and removing the outlier reflection points from the candidate reflection point set to obtain the target point set. Thus, quality screening based on signal stability and moving target confidence, combined with outlier removal based on radial velocity, further eliminates noise and interference from outliers on velocity estimation.

[0011] In one possible implementation of the first aspect, the first velocity condition is: the absolute deviation of the reflection point is greater than the median of the absolute deviations of all reflection points in the candidate reflection point set, where the absolute deviation is the deviation between the radial velocity of the reflection point and the median radial velocity of the reflection points in the candidate reflection point set. Thus, using the absolute median difference statistical method for outlier removal can adaptively identify and eliminate abnormal reflection points in the velocity distribution that deviate significantly from the median, exhibiting strong robustness.

[0012] In one possible implementation of the first aspect, the point cloud error includes at least one of the following: lateral distance error, longitudinal distance error, radial velocity error, lateral velocity error, longitudinal velocity error, azimuth error, and heading error.

[0013] In a second aspect, this application provides an electronic device, including: a memory and a processor; the memory is used to store instructions executed by one or more processors of the electronic device, and the processor is one of one or more processors of the electronic device, used to execute the method in the first aspect or any possible implementation of the first aspect.

[0014] Thirdly, this application provides a readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the method of the first aspect or any possible implementation thereof.

[0015] Fourthly, this application provides a computer program product, which includes computer instructions that, when executed by an electronic device, cause the method in the first aspect or any possible implementation of the first aspect to be implemented.

[0016] The beneficial effects of the second to fourth aspects are similar to those of the first aspect and will not be elaborated here. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a radar point cloud error assessment method is shown according to an embodiment of this application;

[0018] Figure 2 A flowchart illustrating another radar point cloud error assessment method is shown according to an embodiment of this application;

[0019] Figure 3 A schematic diagram of the hardware structure of an electronic device is shown according to an embodiment of this application. Detailed Implementation

[0020] To facilitate understanding of the embodiments of this application, the following describes the three directions involved in this application:

[0021] Longitudinal: refers to the direction of travel of the vehicle. In this application, longitudinal distance and longitudinal speed are based on this direction.

[0022] Lateral: refers to the direction perpendicular to the vehicle's direction of travel in the horizontal plane. In this application, lateral distance and lateral speed are based on this direction.

[0023] Radial: refers to the direction of the line connecting the vehicle and the reflection point, i.e., the radar's line-of-sight direction. The speed directly measured by millimeter-wave radar is the radial speed. In this application, radial speed and radial distance are based on this direction.

[0024] The illustrative embodiments of this application include, but are not limited to, a radar point cloud error assessment method, device, medium, and product.

[0025] As previously mentioned, in some embodiments, the mean or median value of the longitudinal velocities of multiple reflection points within the target detection box is taken as the estimated value of the target vehicle's speed, and then compared with the true value to evaluate the point cloud error. However, in close-range scenarios, the target vehicle's reflection points are densely packed and the longitudinal velocities of each reflection point differ significantly. The statistical mean or median value is prone to deviating from the true speed of the target vehicle, resulting in inaccurate error evaluation results.

[0026] To address the aforementioned issues, this application provides a radar point cloud error assessment method, comprising: acquiring the true values ​​of motion parameters of the target vehicle relative to the vehicle itself; acquiring point cloud data, which is data collected by millimeter-wave radar containing multiple reflection points; acquiring a detection box of the target vehicle labeled based on the point cloud data, the detection box being used to delineate the range of reflection points belonging to the target vehicle in the point cloud data; when the longitudinal distance between the center of the detection box and the vehicle is less than or equal to a first distance threshold, determining the reflection points within the detection box whose absolute value of the azimuth angle relative to the vehicle satisfies a first angle condition as candidate points, and determining the reflection points among the candidate points whose point cloud intensity satisfies a first intensity condition as target points, wherein the longitudinal direction is the driving direction of the vehicle; comparing the estimated values ​​of the motion parameters of the target points in the point cloud data with the true values ​​of the motion parameters to determine the point cloud error.

[0027] Through the embodiments of this application, under close-range conditions, target points are screened by angle conditions and intensity conditions. The angle-based screening can retain the reflection point that best represents the overall motion of the target vehicle in terms of spatial location, while the intensity-based screening can select the most reliable reflection point in terms of signal quality. Thus, representative and reliable target points are screened for point cloud error assessment, which can improve the accuracy and reliability of point cloud error assessment.

[0028] It is understandable that the larger the absolute value of the azimuth angle, the further the reflection point deviates from the line connecting the vehicle and the target vehicle, and the more it is located in the edge area of ​​the target vehicle. It is more susceptible to local motion and scattering interference, and its longitudinal velocity is less representative of the overall motion of the target vehicle. Conversely, the smaller the absolute value of the azimuth angle, the closer the reflection point is to the center area of ​​the target vehicle on the side facing the vehicle, and its longitudinal velocity is more representative of the overall motion of the target vehicle.

[0029] For example, the first angle condition is to select the N reflection points with the smallest absolute value of the direction angle as candidate points, so that the reflection points that best represent the overall motion of the target vehicle can be retained in space.

[0030] It is understandable that the greater the point cloud intensity, the stronger the signal and the higher the signal-to-noise ratio at the reflection point, and the more reliable the measurement results; conversely, the smaller the point cloud intensity, the weaker the signal, or the more likely noise or clutter may be mixed in.

[0031] For example, the first intensity condition is to select the point with the highest reflection intensity in the point cloud from the candidate points as the target point, so as to further select the point with the most reliable signal from the candidate points with the best spatial location, thus ensuring the accuracy of velocity estimation.

[0032] According to some embodiments of this application, the radar point cloud error assessment method can be applied to various electronic devices, such as in-vehicle computing platforms, domain controllers, vehicle-mounted systems, smartphones, tablets, laptops, servers, or other terminal devices or cloud devices with data processing capabilities. This application does not limit the specific type of electronic device.

[0033] The following is based on Figure 1 An embodiment of a radar point cloud error assessment method is introduced.

[0034] like Figure 1 As shown, it includes the following steps:

[0035] S101: Acquire target label data, real-time kinematic (RTK) data of the vehicle, target vehicle RTK data, and point cloud result file. The target label data includes the detection bounding box information of the target vehicle and the first timestamp. The vehicle RTK data and the target vehicle RTK data are data corresponding to the first timestamp obtained by nearest neighbor time matching based on the first timestamp. The point cloud result file includes point cloud data of multiple reflection points corresponding to the first timestamp obtained by time matching based on the first timestamp.

[0036] According to some embodiments, target label data is parsed from a preset input directory. This data includes a timestamp, the center coordinates (X and Y coordinates), width, length, and heading angle of the target vehicle's detection box. The detection box is used to delineate the range of reflection points belonging to the target vehicle in the point cloud data. Based on the parsed detection box center coordinates, width, length, and heading information, the system constructs a region of interest (ROI) in the read point cloud data, obtaining the detection box. This detection box is used to delineate the range of reflection points belonging to the target vehicle in the point cloud data of that frame, and to filter out reflection points falling within this region.

[0037] Among them, the X-coordinate of the center of the detection frame of the target vehicle is used as the estimated value of the target's lateral distance, and the Y-coordinate is used as the estimated value of the target's longitudinal distance. The estimated value of the target's azimuth angle (the angle of the target relative to the direction of travel of the vehicle) is calculated from the XY coordinates. The heading angle is taken from the heading angle field in the target label data and is used as the estimated value of the target's heading angle (the orientation of the target vehicle itself).

[0038] The target label data can be manually labeled or the prediction results of detection boxes generated by automatic labeling algorithms.

[0039] According to some embodiments, before obtaining the detection box, the target label data also needs to be cleaned to remove abnormal detection box data.

[0040] According to some embodiments, RTK data of the vehicle and the target vehicle are read from a preset input directory, supporting multiple text formats such as .txt, .dat, and .rtk. After parsing, the timestamp, latitude and longitude, heading angle, and northbound and eastbound speed fields are uniformly output. For the RTK timestamp, if it contains global positioning system (GPS) weeks, it is converted to a Unix nanosecond timestamp; for the heading angle, it is uniformly normalized to the interval [-180, 180). The parsing results are sorted by timestamp, generating sorted RTK sequences for the vehicle and the target vehicle. Then, using the timestamp in the target label data (i.e., an example of the first timestamp) as the time anchor, nearest neighbor matching is performed in the RTK sequences of the vehicle and the target vehicle respectively, to obtain the matched RTK data of the vehicle and the target vehicle corresponding to the first timestamp.

[0041] According to some embodiments, the point cloud result file contains information on multiple reflection points, and the information of each point includes fields such as XYZ coordinates, point cloud intensity, radial velocity, consistency index, moving target threshold, and correlation index.

[0042] For example, the point cloud result file (such as a millimeter-wave radar point cloud result file) is named with a timestamp, which includes the point cloud data corresponding to the first timestamp. For instance, the first timestamp in the target label data can be used as a time anchor point to match the point cloud data corresponding to the first timestamp in the point cloud result file.

[0043] S102: Based on the RTK data of the vehicle and the RTK data of the target vehicle, obtain the true values ​​of the motion parameters of the target vehicle relative to the vehicle.

[0044] It is understandable that the true values ​​of the motion parameters of the target vehicle relative to the own vehicle represent the motion parameter values ​​of the target vehicle with the own vehicle as a reference. For example, the true values ​​of motion parameters include one or more of the following: the true value of the lateral distance of the target vehicle relative to the own vehicle (representing the true distance of the target vehicle in the left-right direction of the own vehicle), the true value of the longitudinal distance of the target vehicle relative to the own vehicle (representing the true distance of the target vehicle in the forward direction of the own vehicle), the true value of the azimuth angle of the target vehicle relative to the own vehicle (representing the angle of the target vehicle relative to the forward direction of the own vehicle), the true value of the heading angle of the target vehicle (representing the orientation of the target vehicle itself), the true value of the radial velocity of the target vehicle (representing the true velocity of the target vehicle along the line connecting the own vehicle and the target vehicle), the true value of the lateral velocity of the target vehicle (representing the true velocity of the target vehicle in the left-right direction of the own vehicle), and the true value of the longitudinal velocity of the target vehicle (representing the true velocity of the target vehicle in the forward direction of the own vehicle).

[0045] According to some embodiments, the true values ​​of motion parameters can be calculated based on dual-vehicle RTK data. Dual-vehicle RTK data includes the RTK data of the vehicle itself and the RTK data of the target vehicle. The RTK data is acquired through joint calculation using a Global Navigation Satellite System (GNSS) and differential reference stations. For example, the RTK data of the vehicle itself includes one or more of the following: timestamp, latitude and longitude coordinates, northward velocity, eastward velocity, and azimuth angle. The RTK data of the target vehicle includes one or more of the following: timestamp, latitude and longitude coordinates, northward velocity, eastward velocity, and azimuth angle.

[0046] S103: When the longitudinal distance between the center of the detection frame and the vehicle is less than or equal to the first distance threshold, based on the point cloud data of the reflection points in the detection frame, the reflection points in the detection frame whose absolute azimuth angle relative to the vehicle satisfies the first angle condition are determined as candidate points, and the candidate points whose point cloud intensity satisfies the first intensity condition are determined as target points, where the longitudinal direction is the driving direction of the vehicle.

[0047] According to some embodiments, the first distance threshold can be set according to actual needs, for example, 10.0 meters.

[0048] For example, the first angle condition is that the absolute value of the azimuth angle is among the N smallest absolute values ​​of the azimuth angles among all reflection points within the detection box, where N is a positive integer, such as 5. The first intensity condition is that the point cloud intensity is the highest among multiple candidate points.

[0049] According to some embodiments, candidate points can be filtered based on height range before being determined. For example, reflection points with heights in the range of [0.2, 2.0] meters can be filtered to remove ground clutter and excessively high reflection points.

[0050] S104: Compare the estimated motion parameters of the target points in the point cloud data with the true values ​​of the motion parameters to determine the point cloud error.

[0051] According to some embodiments, the estimated values ​​of each motion parameter can be paired with the true values ​​of each motion parameter and the point cloud errors can be calculated.

[0052] For example, the motion parameter estimates include the radial velocity estimate of the target point, the lateral velocity estimate and longitudinal velocity estimate obtained by decomposing the radial velocity estimate, and the azimuth angle estimate and heading angle estimate extracted from the target label data. For example, after the target point is determined in S103, the radial velocity of the target point can be extracted as the radial velocity estimate of the target vehicle. Then, based on the heading angle estimate of the target detection box, the radial velocity estimate is decomposed into lateral velocity estimates and longitudinal velocity estimates. For example, using cos(azimuth angle - heading angle), the radial velocity is decomposed into lateral velocity estimates and longitudinal velocity estimates based on the azimuth angle and heading angle estimates of the target detection box.

[0053] The true values ​​of motion parameters refer to the actual motion parameters of the target vehicle relative to the vehicle obtained in S101, such as at least one of the following: true values ​​of lateral distance, true values ​​of longitudinal distance, true values ​​of azimuth angle, true values ​​of heading angle, true values ​​of longitudinal velocity, true values ​​of lateral velocity, and true values ​​of radial velocity.

[0054] According to some embodiments, the estimated values ​​of each motion parameter can be paired with the corresponding true values ​​of each motion parameter, and various point cloud errors can be calculated. For example, point cloud errors may include at least one of the following: lateral distance error, longitudinal distance error, longitudinal velocity error, lateral velocity error, longitudinal velocity error, azimuth error, and heading error.

[0055] According to some embodiments, the calculation method of each point cloud error is explained in Table 1 below.

[0056] Table 1. Calculation Methods for Point Cloud Error

[0057]

[0058] According to some implementations, the 2σ standard deviation can be calculated for each point cloud error as a statistical output.

[0059] In this embodiment, in close-range scenarios, the N reflection points with the smallest absolute azimuth angle within the detection box are selected based on the azimuth angle condition. Reflection points near the center region of the target vehicle's side facing the vehicle are retained, avoiding interference from edge reflection points on velocity estimation. Then, the reflection point with the highest point cloud intensity is selected as the target point based on the intensity condition. The angle and intensity conditions work together to filter reflection points sequentially, ultimately using the longitudinal velocity of a single target point to represent the target vehicle's velocity. This overcomes the problem of accurately estimating target velocity in close-range scenarios due to the dense reflection points and discrete longitudinal velocities, improving the accuracy and reliability of point cloud error assessment.

[0060] The following is based on Figure 2 An embodiment of another radar point cloud error assessment method is introduced.

[0061] Reference Figure 2 , Figure 2 A flowchart illustrating another radar point cloud error assessment method provided in an embodiment of this application is shown. Figure 2 The process shown includes steps S201 to S207. Specifically, S201-S202 and... Figure 1 S101 corresponds to S203 and Figure 1 S102 corresponds to S204-S205a and Figure 1 S103 corresponds to S206 and Figure 1 S104 corresponds to; S205b is Figure 2 The newly added long-distance scene processing step, S207, is... Figure 2 The newly added statistical output step.

[0062] S201: Obtain target label data, which includes information about the target vehicle's detection frame and its first timestamp.

[0063] According to some embodiments, target label data is parsed from a preset input directory. This data includes a timestamp, the center coordinates (X and Y coordinates), width, length, and heading angle of the target vehicle's detection box. The detection box is used to delineate the range of reflection points belonging to the target vehicle in the point cloud data. Based on the parsed center coordinates, width, length, and heading information of the detection box, the system constructs a region of interest in the read point cloud data, thus obtaining the detection box.

[0064] Among them, the X-coordinate of the center of the detection frame of the target vehicle is used as the estimated value of the target's lateral distance, and the Y-coordinate is used as the estimated value of the target's longitudinal distance. The estimated value of the target's azimuth angle (the angle of the target relative to the direction of travel of the vehicle) is calculated from the XY coordinates. The heading angle is taken from the heading angle field in the target label data and is used as the estimated value of the target's heading angle (the orientation of the target vehicle itself).

[0065] The target label data can be manually labeled or the prediction results of detection boxes generated by automatic labeling algorithms.

[0066] According to some embodiments, before acquiring the detection frames, the target label data needs to be cleaned to remove abnormal detection frame data. Specifically, the target label data is parsed from a preset input directory, extracting the timestamp, the center coordinates (X and Y coordinates), width, length, and heading angle information of the target vehicle's detection frame, arranged in chronological order, and then label cleaning is performed. For example, cleaning the target label data includes detecting anomalies such as jumps in the center position of the target vehicle's detection frame. If the change in the center position of a target vehicle's detection frame falls within a preset threshold range, the detection frame of that target vehicle is determined to be abnormal data and is removed and cleaned.

[0067] S202: Obtain the real-time dynamic differential positioning (RTK) data of the vehicle, the RTK data of the target vehicle, and the point cloud result file. Using the first timestamp as the time anchor, perform time matching on the vehicle RTK data, the target vehicle RTK data, and the point cloud result file respectively to obtain the vehicle RTK data, the target vehicle RTK data, and the point cloud data corresponding to the first timestamp.

[0068] According to some embodiments, RTK data of the vehicle and the target vehicle are read from a preset input directory, supporting multiple text formats such as .txt, .dat, and .rtk. After parsing, the timestamp, latitude and longitude, heading angle, and northbound and eastbound speed fields are uniformly output. For the RTK timestamp, if it contains GPS weeks, it is converted to a Unix nanosecond timestamp; for the heading angle, it is uniformly normalized to the interval [-180, 180). The parsing results are sorted by timestamp, generating sorted RTK sequences for the vehicle and the target vehicle. Then, using the timestamp in the target tag data (i.e., the first timestamp) as the time anchor point, nearest neighbor matching is performed in both the RTK sequences for the vehicle and the target vehicle, respectively, to obtain the matched RTK data for the vehicle and the target vehicle corresponding to the first timestamp.

[0069] According to some embodiments, the point cloud result file is named with a timestamp and includes information on multiple reflection points. The information of each reflection point includes fields such as XYZ coordinates, point cloud intensity, radial velocity, consistency index, moving target threshold, and correlation index.

[0070] S203: Based on the matched RTK data of the self-vehicle and the RTK data of the target vehicle, obtain the true values ​​of the motion parameters of the target vehicle relative to the self-vehicle.

[0071] It is understandable that the true values ​​of the motion parameters of the target vehicle relative to the self-vehicle represent the motion parameter values ​​of the target vehicle with the self-vehicle as a reference. For example, the true values ​​of motion parameters include one or more of the following: the true value of the lateral distance of the target vehicle relative to the self-vehicle, the true value of the longitudinal distance of the target vehicle relative to the self-vehicle, the true value of the azimuth angle of the target vehicle relative to the self-vehicle, the true value of the heading angle of the target vehicle, the true value of the radial velocity of the target vehicle, the true value of the lateral velocity of the target vehicle, and the true value of the longitudinal velocity of the target vehicle.

[0072] According to some embodiments, the true values ​​of the motion parameters of the target vehicle relative to the self-vehicle include the true values ​​of lateral distance, longitudinal distance, azimuth angle, heading angle, longitudinal velocity, lateral velocity, and radial velocity.

[0073] The following describes an embodiment for calculating the true values ​​of the lateral distance, longitudinal distance, and azimuth angle of the target relative to the vehicle. For example, the latitude and longitude coordinates in both the vehicle's RTK data and the target vehicle's RTK data are transformed to the vehicle's coordinate system. The difference between the two coordinates yields the true values ​​of the lateral distance, longitudinal distance, and azimuth angle of the target relative to the vehicle. Since there is a fixed offset between the RTK antenna mounting point and the vehicle's center reference point, the RTK output position is the position of the antenna mounting point, not the position of the vehicle's center. Therefore, the true value of the longitudinal distance is compensated based on the fixed longitudinal offset between the RTK antenna mounting point and the vehicle's center. The compensated true value of the longitudinal distance is equivalent to the longitudinal distance relative to the vehicle's center.

[0074] For example, the true value of the heading angle of the target vehicle can be obtained directly from the heading angle field of the target vehicle's RTK data.

[0075] The following describes an embodiment for calculating the true values ​​of the relative lateral velocity, longitudinal velocity, and radial velocity of the target vehicle in the driver's coordinate system. For example, based on the northward and eastward velocities of the driver and the target vehicle, the driver's heading, the target vehicle's heading, and the target's azimuth, the true values ​​of the relative lateral velocity, longitudinal velocity, and radial velocity of the target vehicle in the driver's coordinate system are calculated.

[0076] S204: Determine the relationship between the longitudinal distance between the center of the detection frame and the vehicle and the first distance threshold.

[0077] For example, the longitudinal distance between the center of the detection frame and the vehicle is determined based on the center coordinates of the detection frame of the target vehicle in the target label data.

[0078] If the longitudinal distance between the center of the detection frame and the vehicle is less than or equal to the first distance threshold (e.g., 10 meters), then execute S205a; if the longitudinal distance is greater than the first distance threshold, then execute S205b.

[0079] According to some embodiments, the execution judgment of branches S205a-S206a and S205b-S206b is not limited to being based on a fixed threshold. For example, hysteresis interval switching or continuous weight fusion switching can also be used.

[0080] Hysteresis interval switching refers to setting two different distance thresholds. When the target vehicle moves from far to near into the close-range scene, a smaller threshold is used, and when the target vehicle moves from near to far out of the close-range scene, a larger threshold is used. The area between the two thresholds is the hysteresis interval. Within this interval, the original processing method is maintained without switching, so as to avoid frequent switching near the threshold.

[0081] Continuous weighted fusion switching refers to the fusion of speed estimates obtained from the near-distance path and the far-distance path in a transition interval near a threshold, weighted according to distance. The closer the distance, the greater the weight of the near-distance path result, and the farther the distance, the greater the weight of the far-distance path result, so as to make the speed estimate transition smoothly at the switching boundary.

[0082] S205a: When the longitudinal distance between the center of the detection frame and the vehicle is less than or equal to the first distance threshold, the reflection point within the detection frame whose absolute azimuth angle relative to the vehicle satisfies the first angle condition is determined as a candidate point, and the candidate point whose point cloud intensity satisfies the first intensity condition is determined as the target point, where the longitudinal direction is the driving direction of the vehicle.

[0083] According to some embodiments, the first distance threshold can be set according to actual needs, for example, 10.0 meters.

[0084] For example, the first angle condition is that the absolute value of the azimuth angle is among the N smallest absolute values ​​of the azimuth angles among all reflection points within the detection box, where N is a positive integer, such as 5. The first intensity condition is that the point cloud intensity is the highest among multiple candidate points.

[0085] According to some embodiments, candidate points can be filtered based on height range before being determined. For example, reflection points with heights in the range of [0.2, 2.0] meters can be filtered to remove ground clutter and excessively high reflection points.

[0086] S206a: Compare the estimated motion parameters of the target point with the true motion parameters to determine the point cloud error.

[0087] According to some embodiments, after the target point is determined in S205a, the radial velocity of the target point is extracted as the speed estimate of the target vehicle, and based on the heading angle estimate of the target detection box, the radial velocity is decomposed into a lateral velocity estimate and a longitudinal velocity estimate. For example, cos(azimuth angle - heading angle) is used to decompose the radial velocity into lateral and longitudinal components to obtain the lateral velocity estimate and the longitudinal velocity estimate.

[0088] The true values ​​of motion parameters refer to the actual motion parameters of the target vehicle relative to the vehicle itself obtained in S201, including the true values ​​of lateral distance, longitudinal distance, azimuth angle, heading angle, longitudinal velocity, lateral velocity, and radial velocity.

[0089] The estimated values ​​of each motion parameter are paired with their corresponding true values, and various point cloud errors are calculated. For example, point cloud errors may include at least one of the following: lateral distance error, longitudinal distance error, longitudinal velocity error, lateral velocity error, longitudinal velocity error, azimuth error, and heading error.

[0090] S205b: When the longitudinal distance between the center of the detection frame and the vehicle is greater than the first distance threshold, the reflection points within the detection frame are subjected to quality screening and outlier removal to obtain the target point set.

[0091] According to some embodiments, reflection points within the detection frame are filtered based on the signal stability of each reflection point and the confidence level that it belongs to a moving target, resulting in a candidate reflection point set. For example, filtering is performed based on a consistency index and a moving target threshold, retaining reflection points with a consistency ≥ 0.2 and a moving target threshold ≥ 0.2.

[0092] According to some embodiments, outlier reflection points whose radial velocities do not meet a first velocity condition are determined based on the radial velocities of each reflection point in the candidate reflection point set. The first velocity condition is that the absolute deviation of the reflection point is greater than the median of the absolute deviations of all reflection points in the candidate reflection point set, where the absolute deviation is the deviation between the radial velocity of the reflection point and the median radial velocity of the reflection points in the candidate reflection point set. It can be understood that radial direction refers to the direction of the line connecting the vehicle and the reflection point. By removing outlier reflection points from the candidate reflection point set, the target point set is obtained. This application does not limit the method of outlier removal. For example, the outlier removal method can be any one of the following: median absolute deviation (MAD) statistical method, interquartile range (IQR) statistical method, quantile statistical method, random sampling consensus (RANSAC) statistical method, truncated mean statistical method, or Kalman smoothing statistical method.

[0093] S206b: Compare the average estimated value of the motion parameters of the target point set in the point cloud data with the true value of the motion parameters to obtain the point cloud error.

[0094] According to some embodiments, the radial velocity of each reflection point in the target point set is averaged as the velocity estimate of the target vehicle. Based on the heading angle estimate of the target detection box, the radial velocity is decomposed into lateral velocity estimate and longitudinal velocity estimate. For example, cos(azimuth angle - heading angle) is used to decompose the radial velocity into lateral and longitudinal components as lateral velocity estimate and longitudinal velocity estimate.

[0095] It can be understood that the true values ​​of motion parameters refer to the actual motion parameters of the target vehicle relative to the self-vehicle obtained in S201, including the true values ​​of lateral distance, longitudinal distance, azimuth angle, heading angle, longitudinal velocity, lateral velocity, and radial velocity.

[0096] According to some embodiments, the average estimated values ​​of motion parameters of the target point set are paired with the corresponding true values ​​of motion parameters, and various point cloud errors are calculated based on the paired average estimated values ​​of motion parameters and the true values ​​of motion parameters. For example, point cloud errors may include at least one of lateral distance error, longitudinal distance error, longitudinal velocity error, lateral velocity error, longitudinal velocity error, azimuth error, and heading error.

[0097] S207: Output statistical evaluation indicators.

[0098] After error calculation, statistical evaluation indicators are output based on the error calculation results.

[0099] For example, the standard deviations of the lateral distance error, longitudinal distance error, azimuth angle error, heading angle error, longitudinal velocity error, lateral velocity error and radial velocity error in the point cloud error are calculated respectively, and twice the standard deviation is taken as the global statistical index.

[0100] For example, based on the absolute value of the azimuth angle of the detection frame center relative to the vehicle, the point cloud error data corresponding to each target vehicle is divided into different azimuth angle groups. For instance, when |azimuth angle| ≤ 15°, the lateral velocity error and longitudinal velocity error of the target vehicle are assigned to the first azimuth angle group; when |azimuth angle| > 15°, the lateral velocity error and longitudinal velocity error of the target vehicle are assigned to the second azimuth angle group. The 2σ standard deviation of the lateral velocity error and longitudinal velocity error within each of the two groups is calculated respectively.

[0101] For example, the point cloud error data corresponding to each target vehicle is grouped according to the longitudinal distance between the center of the detection frame and the vehicle. For instance, the distance grouping intervals are: 0-10m, 10-30m, 30-50m, 50-70m, 70-100m, 100-150m, 150-200m, and greater than 200m. The mean distance error and the mean square error of the velocity error are calculated for each distance group. For example, if the longitudinal distance between the center of the detection frame of target vehicle A and the vehicle is 25 meters, then the distance error and velocity error corresponding to this target vehicle are assigned to the 10-30m distance group and included in the calculation of the mean distance error and the mean square error of the velocity error for that group.

[0102] In this embodiment, the RTK data of the self-vehicle and the target vehicle are aligned under the timestamp of the target label data by the dual-vehicle RTK nearest neighbor matching, which solves the problem of time asynchrony of multi-source data; by transforming the latitude and longitude coordinates to the self-vehicle coordinate system, the true value of RTK is consistent with the horizontal and vertical definitions of the millimeter-wave radar point cloud output, and the fixed vertical offset between the RTK installation point and the vehicle center is compensated, eliminating the systematic distance deviation.

[0103] Furthermore, based on the relationship between the longitudinal distance between the center of the detection box and the vehicle and the first distance threshold, the system adaptively switches between near-field and far-field processing paths. In near-field scenarios, reflection points within the detection box closest to the center of the target vehicle are selected using azimuth angle conditions, and then the most reliable target point is chosen from these based on intensity conditions. This effectively avoids the mean estimation bias caused by dense reflection points and discrete velocity in near-field scenarios. In far-field scenarios, reflection points are quality-screened using consistency indices and moving target thresholds, and velocity outliers are removed using the MAD absolute median difference statistical method. This effectively suppresses interference from sparse noise and outliers in far-field scenarios. Through these methods, the overall stability of point cloud velocity estimation is enhanced.

[0104] Furthermore, in this embodiment, not only can the global 2σ standard deviation be output, but the error can also be grouped and statistically analyzed according to azimuth angle (e.g., within and outside ±15°) and distance (e.g., multiple intervals such as 0-10m, 10-30m, etc.). Through this multi-level error analysis, the distribution characteristics of the error at different distances and angles can be obtained, providing a more refined evaluation basis for radar calibration, algorithm acceptance, and problem localization.

[0105] further, Figure 3 According to some embodiments of this application, a schematic diagram of the structure of an electronic device 100 is shown. For example... Figure 3As shown, the electronic device 100 includes one or more processors 101, system memory 102, non-volatile memory (NVM) 103, communication interface 104, input / output (I / O) device 105, and system control logic 106 for coupling the processor 101, system memory 102, non-volatile memory 103, communication interface 104 and input / output (I / O) device 105.

[0106] Wherein, processor 101 may include one or more processing units, such as processing modules or processing circuits that may include central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), microprocessors (MCUs), AI (Artificial Intelligence) processors, or programmable logic devices (FPGAs), neural network processing units (NPUs), etc., and may include one or more single-core or multi-core processors. According to some embodiments, the processor is a superscalar processor.

[0107] System memory 102 is volatile memory, such as random-access memory (RAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc. System memory is used for temporary storage of data and / or instructions.

[0108] The non-volatile memory 103 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. In some embodiments, the non-volatile memory 103 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device, such as a hard disk drive (HDD), a compact disc (CD), a digital versatile disc (DVD), a solid-state drive (SSD), etc. In some embodiments, the non-volatile memory 103 may also be a removable storage medium, such as a Secure Digital (SD) memory card, etc.

[0109] Specifically, system memory 102 and non-volatile memory 103 may each include a temporary copy and a permanent copy of instruction 107. Instruction 107 may include, when executed by at least one of processors 101, causing electronic device 100 to implement the data processing methods provided in the embodiments of this application.

[0110] The communication interface 104 may include a transceiver for providing a wired or wireless communication interface for the electronic device 100, thereby enabling communication with any other suitable device via one or more networks. In some embodiments, the communication interface 104 may be integrated into other components of the electronic device 100, for example, the communication interface 104 may be integrated into the processor 101. In some embodiments, the electronic device 100 may communicate with other devices through the communication interface 104.

[0111] Input / output (I / O) device 105 can be an input device such as a keyboard or mouse, and an output device such as a monitor. Users can interact with electronic device 100 through input / output (I / O) device 105, such as inputting a neural network model to be run.

[0112] System control logic 106 may include any suitable interface controller to provide any suitable interface to other modules of electronic device 100. For example, in some embodiments, system control logic 106 may include one or more memory controllers to provide an interface to system memory 102 and non-volatile memory 103.

[0113] In some embodiments, at least one of the processors 101 may be packaged together with the logic of one or more controllers for system control logic 106 to form a system in package (SiP). In other embodiments, at least one of the processors 101 may also be integrated on the same chip with the logic of one or more controllers for system control logic 106 to form a system-on-chip (SoC).

[0114] Understandable. Figure 3 The structure of the electronic device 100 shown is merely an example. In other embodiments, the electronic device 100 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0115] It is understood that electronic device 100 can be any electronic device, including but not limited to mobile phones, wearable devices (such as smartwatches), tablets, desktops, laptops, handheld computers, notebook computers, ultra-mobile personal computers (UMPCs), netbooks, as well as cellular phones, personal digital assistants (PDAs), augmented reality (AR) / virtual reality (VR) devices, etc., which are not limited in the embodiments of this application.

[0116] Various embodiments of the mechanisms disclosed in this application can be implemented in hardware, software, firmware, or combinations of these implementation methods. Embodiments of this application can be implemented as computer programs or program code executable on a programmable system, the programmable system including at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0117] Program code can be applied to input instructions to execute the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, the processing system includes any system having a processor such as, for example, a Digital Signal Processor (DSP), a microcontroller, an Application Specific Integrated Circuit (ASIC), or a microprocessor.

[0118] The program code can be implemented using a high-level procedural language or an object-oriented programming language to communicate with the processing system. Assembly language or machine language can also be used when needed. In fact, the mechanisms described in this application are not limited to any particular programming language. In either case, the language can be a compiled language or an interpreted language.

[0119] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored thereon on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, the instructions may be distributed via a network or through other computer-readable media. Therefore, machine-readable media may include any mechanism for storing or transmitting information in a machine-readable (e.g., computer-readable) form, including but not limited to floppy disks, optical disks, CD-ROMs, magneto-optical disks, read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic cards or optical cards, flash memory, or tangible machine-readable storage for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) using the Internet in the form of electrical, optical, acoustic, or other forms of propagated signals. Therefore, machine-readable media include any type of machine-readable medium suitable for storing or transmitting electronic instructions or information in a machine-readable (e.g., computer-readable) form.

[0120] This application also provides a readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the methods provided in the above embodiments.

[0121] This application also provides a computer program product, which includes computer instructions that, when executed by an electronic device, cause the methods provided in the above embodiments to be implemented.

[0122] This application also provides an electronic device, which includes a memory and a processor. The memory is used to store instructions executed by one or more processors of the electronic device. The processor is one of the one or more processors of the electronic device and is used to execute the instructions to implement the methods provided in the above embodiments.

[0123] In the accompanying drawings, some structural or methodological features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be necessary. Rather, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Furthermore, the inclusion of structural or methodological features in a particular figure does not imply that such features are required in all embodiments, and in some embodiments, these features may be omitted or may be combined with other features.

[0124] It should be noted that all units / modules mentioned in the device embodiments of this application are logical units / modules. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or a combination of multiple physical units / modules. The physical implementation of these logical units / modules themselves is not the most important factor; the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. Furthermore, to highlight the innovative aspects of this application, the above-described device embodiments of this application have not introduced units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that the above-described device embodiments do not contain other units / modules.

[0125] It should be noted that, in the examples and specification of this patent, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0126] Although this application has been illustrated and described with reference to certain preferred embodiments thereof, those skilled in the art should understand that various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A radar point cloud error assessment method, characterized in that, include: The system acquires target label data, vehicle RTK data, target vehicle RTK data, and a point cloud result file. The target label data includes the target vehicle's detection bounding box information and a first timestamp. The vehicle RTK data and the target vehicle RTK data are respectively data corresponding to the first timestamp obtained through nearest neighbor time matching based on the first timestamp. The point cloud result file includes point cloud data of multiple reflection points corresponding to the first timestamp, obtained through time matching based on the first timestamp. Based on the RTK data of the vehicle and the RTK data of the target vehicle, obtain the true values ​​of the motion parameters of the target vehicle relative to the vehicle. When the longitudinal distance between the center of the detection frame and the vehicle is less than or equal to a first distance threshold, based on the point cloud data of the reflection points within the detection frame, the reflection points within the detection frame whose absolute azimuth angle relative to the vehicle satisfies the first angle condition are determined as candidate points, and the candidate points whose point cloud intensity satisfies the first intensity condition are determined as target points, wherein the longitudinal direction is the driving direction of the vehicle; The estimated motion parameters of the target point in the point cloud data are compared with the true values ​​of the motion parameters to determine the point cloud error.

2. The method according to claim 1, characterized in that, The first angle condition is: The absolute azimuth angle is the N smallest absolute azimuth angle values ​​among all reflection points within the detection frame, where N is a positive integer.

3. The method according to claim 1, characterized in that, The number of candidate points is multiple. The first intensity condition is: the point cloud intensity is the highest among the multiple candidate points.

4. The method according to claim 1, characterized in that, The method further includes: If the longitudinal distance between the center of the detection frame and the vehicle is greater than the first distance threshold, the reflection points within the detection frame are subjected to quality screening and outlier removal to determine the target point set. The average estimated value of the motion parameters of the target point set in the point cloud data is compared with the true value of the motion parameters to determine the point cloud error.

5. The method according to claim 4, characterized in that, The step of performing quality screening and outlier removal on the reflection points within the detection frame to determine the target point set includes: Based on the signal stability of each reflection point within the detection frame and the confidence level that it belongs to a moving target, the reflection points within the detection frame are filtered to obtain a candidate reflection point set; Based on the radial velocity of each reflection point in the candidate reflection point set, outlier reflection points whose radial velocity does not satisfy the first velocity condition are determined. The outlier reflection points are removed from the candidate reflection point set to obtain the target point set.

6. The method according to claim 5, characterized in that, The first speed condition is: The absolute deviation of the reflection point is greater than the median value of the absolute deviations of all reflection points in the candidate reflection point set. Wherein, the absolute deviation is the deviation between the radial velocity of the reflection point and the median value of the radial velocity of the concentrated reflection points of the candidate reflection points.

7. The method according to any one of claims 1-6, characterized in that, The point cloud error includes at least one of the following: Lateral distance error, longitudinal distance error, radial velocity error, lateral velocity error, longitudinal velocity error, azimuth error, heading error.

8. An electronic device, characterized in that, include: Memory, processor; The memory is used to store instructions executed by one or more processors of the electronic device, the processor being one of the one or more processors of the electronic device, for performing the method according to any one of claims 1 to 7.

9. A readable storage medium, characterized in that, The readable storage medium stores instructions that, when executed on an electronic device, cause the electronic device to perform the method of any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes computer instructions that, when executed by an electronic device, cause the method of any one of claims 1 to 7 to be implemented.

Images