A vehicle-mounted millimeter wave radar error correction method for complex traffic scenes

Abstract

The application discloses a kind of vehicle-mounted millimeter wave radar error correction methods for complex traffic scene, it is related to radar data processing technical field, and stable weak scattering point set Sca is extracted by candidate scattering point set Csp, and the weak point trace in the sheltered area is not ignored by the aid of boundary corresponding set Bdr, so that track keeps continuity and credibility on time series and spatial coordinates;Combined with the track compensation set Com generated by time synchronization error Ets and calibration residual Ecl and applied to track fitting set Fit, the corrected track set Cor obtained can effectively eliminate the drift and misplacement caused by radar sampling delay and installation angle deviation, to correct track set Cor as output result is used for scene reconstruction and target identification, so that vehicles can still obtain time alignment and spatially accurate error correction final track results in complex traffic environment such as night multi-lane highway, rain and fog weather intersection.

Description

Technical Field

[0001] This invention relates to the field of radar data processing technology, specifically to an error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios. Background Technology

[0002] As intelligent driving and advanced driver assistance systems continue to develop, vehicle-mounted millimeter-wave radar, with its advantages of all-weather operation, long range, and stable perception, has gradually become a core component of vehicle environmental perception systems. Especially in complex traffic scenarios, millimeter-wave radar is widely used in key functions such as forward collision avoidance, lane change assist, and automatic emergency braking, addressing the application needs of dense traffic flow, diverse vehicle types, and complex road environments.

[0003] In existing applications, when large vehicles such as buses and trucks appear in traffic scenarios, they often cause asymmetric occlusion of adjacent small targets, such as motorcycles, bicycles, and even cars. Millimeter-wave radar typically cannot directly detect these occluded small targets in such situations, only acquiring scattered weak scattering points. Most existing technologies treat these weak scattering points as noise data and discard them directly, leading to the loss of target information and creating detection blind spots. This deficiency is particularly severe in complex traffic scenarios such as multi-lane congestion, urban overpasses, and tunnel entrances, easily causing small targets to "disappear" from the perception system in a short time, posing safety hazards when autonomous driving systems perform distance keeping or lane-changing assistance. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios, solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios, comprising the following steps:

[0006] S1. Collect the original radar frame set output by the vehicle-mounted millimeter-wave radar, establish a joint time axis, perform coordinate registration, geometric projection and noise removal on the radar data, and generate the original point cloud dataset Pcl.

[0007] S2. Using the original point cloud dataset Pcl as input, establish a candidate target set Can. Obtain a large target set Big through volume, scale and stability conditions, and generate an occlusion region set Occ. Form a candidate scattering point set Csp within the occlusion region set Occ. Filter the weak scattering point set Sca according to signal strength and stability, and establish a boundary correspondence set Bdr with the large target set Big.

[0008] S3. Based on the weak scattering point set Sca, group the trajectory candidate set Trc according to spatial and temporal conditions; perform fitting calculations on the trajectory candidate set Trc to obtain velocity and acceleration estimates, and obtain the trajectory fitting set Fit through residual error screening.

[0009] S4. Calculate the compensation amount using the time synchronization error Ets and the calibration residual Ecl to form the trajectory compensation amount set Com, and apply it to the trajectory fitting set Fit to obtain the corrected trajectory set Cor;

[0010] S5. The correction trajectory set Cor is used as the final trajectory result for error correction and is directly called in scene reconstruction and target recognition.

[0011] Preferably, S1 includes S11;

[0012] S11. During vehicle operation, the vehicle-mounted millimeter-wave radar collects signals according to a set sampling period, and after receiving, demodulating and initially formatting, outputs the radar raw frame set Rfr.

[0013] The system clock generates a timestamp sequence Tst and establishes a joint time axis Tim, so that the radar's original frame set Rfr and the original motion data sequence Mrd are consistent in the time dimension.

[0014] The alignment difference of the data in the original radar frame set Rfr under the timestamp sequence Tst is calculated to obtain the synchronization offset Dlt, and the synchronization offset Dlt is used to map the data to the joint time axis Tim for timestamp binding.

[0015] Preferably, S1 further includes S12;

[0016] S12. Load the external and internal parameters saved during vehicle installation to form coordinate system calibration parameters Cal, and register the millimeter-wave radar coordinate system with the vehicle body coordinate system through rigid body coordinate transformation.

[0017] Geometric projection is performed on the original radar frame set Rfr based on the coordinate projection mapping method, and noise removal is performed on isolated points and outliers in the point cloud data based on the statistical outlier removal method to generate the original point cloud dataset Pcl.

[0018] Based on the mean square error calculation method, the time synchronization error Ets is obtained by performing difference processing on the joint time axis Tim and the synchronization offset Dlt.

[0019] The original radar frame set Rfr is fitted with the coordinate system calibration parameter Cal using the least squares fitting method to obtain the calibration residual Ecl.

[0020] The joint time axis Tim is statistically processed using a frame coverage statistical method to obtain the data coverage Cov.

[0021] Preferably, S2 includes S21;

[0022] S21. Using the original point cloud dataset Pcl as input, perform density aggregation and time series association on the spatial points to obtain multiple point cloud clusters, and combine these point cloud clusters into a candidate target set Can.

[0023] Within the candidate target set Can, calculate the volume parameter Vol, aspect ratio Rat, and temporal stability Sta for each point cloud cluster;

[0024] The volume parameter Vol is obtained by multiplying the length, width, and height of the three-dimensional bounding box of the point cloud cluster. The calculation form is Vol = length × width × height.

[0025] The aspect ratio Rat is obtained by the ratio of the maximum side length to the minimum side length of the point cloud cluster bounding box, and is calculated as Rat = maximum side length / minimum side length.

[0026] The temporal stability Sta is obtained by statistically analyzing the ratio of the number of times a point cloud cluster appears in consecutive frames to the total number of frames. The calculation form is Sta = number of occurrence frames ÷ total number of frames.

[0027] When the following three conditions are met:

[0028] Volume parameter Vol > volume threshold Vth;

[0029] The aspect ratio Rat falls within the aspect ratio threshold range Rth_min≤Rat≤Rth_max;

[0030] Time stability Sta ≥ stability threshold Sth;

[0031] The point cloud clusters are labeled and then incorporated into the large target set Big.

[0032] Based on the geometric outline of the large target set Big, projection calculations are performed in the vehicle coordinate system to calculate the occluded spatial region, forming the occluded region set Occ, and maintaining the correspondence with the large target set Big.

[0033] Among them, the volume parameter Vol of all point cloud clusters in the candidate target set Can is statistically analyzed, and its mean value is multiplied by a fixed coefficient K.

[0034] Rth_min and Rth_max represent the minimum and maximum aspect ratio thresholds, respectively, which are determined by the geometric features of common large vehicles, including buses and trains, and are used to limit the proportion of point cloud clusters. The specific values ​​are set by the user.

[0035] The stability threshold Sth is set empirically in real-world application scenarios, indicating that a target is considered a stable large target only if it remains present for at least consecutive frames within the stability threshold Sth.

[0036] Preferably, S2 further includes S22;

[0037] S22. Using the occlusion region set Occ and the original point cloud dataset Pcl as input, extract all points that fall within the occlusion region set Occ and combine them to form a candidate scattering point set Csp.

[0038] Within the candidate scattering point set Csp, two parameters are calculated for each point: signal strength Sig and frame occurrence count Fre;

[0039] Among them, the signal strength Sig is determined by the amplitude of the millimeter-wave radar echo signal;

[0040] Frame occurrence count Fre: Obtained by counting the number of times the point appears in consecutive frames;

[0041] Define filtering conditions for the candidate scattering point set Csp, and filter each point trace according to the filtering conditions:

[0042] Filtering condition 1: Signal strength Sig < strength threshold Sth;

[0043] Filtering condition 2: Frame occurrence count Fre ≥ stability threshold Fth;

[0044] Among them, the intensity threshold Sth is the energy threshold used to distinguish between strong and weak scattering points, and the specific value is set by the user; the stability threshold Fth is used to determine whether weak scattering points remain in continuous frames, and the specific value is set by the user.

[0045] When screening conditions one and two are met, the point is marked and included in the established weak scattering point set Sca;

[0046] The three-dimensional coordinates of each point in the weak scattering point set Sca are compared with the geometric outline of the large target set Big.

[0047] When the spatial location of the point trace in the weak scattering point set Sca falls within the boundary extension range of the large target set Big, a mapping relationship between the point trace and the corresponding large target set Big is established in the boundary corresponding set Bdr.

[0048] Preferably, S3 includes S31;

[0049] S31. Using the set of weak scattering points Sca as input, perform joint calculations on the points in the set in three-dimensional coordinates and time series to obtain the spatial distance Dis and time interval Int.

[0050] The spatial distance Dis represents the three-dimensional Euclidean distance between adjacent points; the time interval Int represents the sampling time difference between adjacent points.

[0051] When the spatial distance Dis ≤ the preset distance threshold Dth and the time interval Int ≤ the preset time threshold Tth, the points are grouped into the same trajectory cluster;

[0052] Group all points that meet the conditions to form a candidate trajectory set Trc.

[0053] Preferably, S3 further includes S32;

[0054] S32. Using the trajectory candidate set Trc as input, perform trajectory fitting analysis on each trajectory cluster in the time series dimension to obtain velocity estimate Vel and acceleration estimate Acc.

[0055] The velocity estimate Vel is calculated from the spatial displacement and time interval between adjacent points;

[0056] The acceleration estimate Acc is calculated from the changes in adjacent velocities and the time interval.

[0057] The trajectory curves of the time series positions of the trajectory clusters are calculated based on the least squares fitting method, the fitting results are obtained, and the residual error value Err of each trajectory is recorded.

[0058] The obtained residual error value Err is then compared and filtered with the preset fitting error threshold Eth to obtain the filtered trajectory fitting set Fit;

[0059] The comparison and screening process was conducted in the following ways:

[0060] When the residual error Err ≤ the fitting error threshold Eth, the residual error value Err of the trajectory cluster is recorded in the trajectory fitting set Fit;

[0061] If the residual error Err is greater than the fitting error threshold Eth, then the trajectory cluster is discarded and not included in the trajectory fitting set Fit.

[0062] Preferably, S4 includes S41;

[0063] S41. Convert the time synchronization error Ets into a time compensation value Tco, and match it with each trajectory in the trajectory fitting set Fit, in order to correct the uniform offset of the trajectory on the time axis.

[0064] The calibration residual Ecl is converted into a calibration compensation value Cco and matched with each trajectory in the trajectory fitting set Fit, which is used to correct the trajectory offset in spatial coordinates.

[0065] Establish a correspondence between the time compensation value Tco, the calibration compensation value Cco, and each trajectory in the trajectory fitting set Fit, forming the trajectory compensation quantity set Com.

[0066] Preferably, S4 further includes S42;

[0067] S42. Apply a time compensation value Tco to all trajectories in the trajectory fitting set Fit to obtain the time-aligned trajectory results.

[0068] A calibration compensation value Cco is uniformly applied to all trajectories in the trajectory fitting set Fit to obtain the spatially corrected trajectory results, which are defined as the corrected trajectory set Cor.

[0069] Preferably, S5 includes S51;

[0070] S51. Taking the correction trajectory set Cor as input, it is used as the final trajectory result for subsequent use by the vehicle system. During the scene reconstruction process, the correction trajectory set Cor is directly called. At the same time, the correction trajectory set Cor is provided to the target detection and tracking stage to represent the position and motion state of the dynamic target.

[0071] This invention provides an error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios, which has the following advantages:

[0072] (1) By using the original point cloud dataset Pcl to establish a candidate target set Can and generate an occlusion region set Occ, a stable weak scattering point set Sca can still be extracted from the candidate scattering point set Csp even when large targets such as buses and trucks form occlusions. The boundary correspondence set Bdr is used to ensure that weak traces in the occlusion region are not ignored, thus avoiding the problem of targets being completely lost in tunnel entrances or large vehicle intersections. By establishing the trajectory candidate set Trc and the trajectory fitting set Fit, the trajectory maintains continuity and reliability in time series and spatial coordinates. The trajectory compensation amount set Com generated by combining the time synchronization error Ets and the calibration residual Ecl is applied to the trajectory fitting set Fit. The resulting corrected trajectory set Cor can effectively eliminate the drift and misalignment caused by radar sampling delay and installation angle deviation. Finally, the corrected trajectory set Cor is used as the output result for scene reconstruction and target recognition, so that vehicles can still obtain time-aligned and spatially accurate error-corrected final trajectory results in complex traffic environments such as multi-lane highways at night and intersections in rainy and foggy weather, ensuring that the detection of surrounding vehicles and obstacles is not delayed or offset.

[0073] (2) By calculating the three-dimensional coordinates and time series of the weak scattering point set Sca, the scattered points are aggregated according to the dual conditions of spatial distance Dis and time interval Int, forming multiple trajectory clusters with spatiotemporal continuity. These clusters are combined into a trajectory candidate set Trc, thus organizing the originally irregular point distribution into an analyzable trajectory sequence. Subsequently, the dynamic characteristics of velocity estimation Vel and acceleration estimation Acc are calculated on the trajectory candidate set Trc, and continuous trajectory curves are obtained through the least squares fitting method. At the same time, a high-confidence trajectory fitting set Fit is generated by comparing the residual error value Err with the fitting error threshold Eth. This process not only makes the trajectory continuous and interpretable in time and space, but also effectively eliminates abnormal trajectories through residual constraints, avoiding interference from noisy point clouds on subsequent error correction.

[0074] (3) The time synchronization error Ets is converted into a time compensation value Tco and applied to all trajectories, thereby ensuring that the trajectory timestamps are strictly consistent with the global clock and avoiding overall trajectory drift caused by radar sampling delay. At the same time, the calibration residual Ecl is converted into a calibration compensation value Cco and applied to the trajectory space coordinates, aligning the trajectory results with the vehicle reference coordinate system and eliminating spatial offset caused by sensor installation position or attitude. After the dual compensation process, the resulting set of corrected trajectories Cor is directly transmitted to the vehicle perception and control system for scene reconstruction and target tracking, thereby ensuring that the description of dynamic target position and motion state is no longer subject to basic measurement errors. Attached Figure Description

[0075] Figure 1 This is a schematic diagram illustrating the steps of an error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios according to the present invention. Detailed Implementation

[0076] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0077] Example 1

[0078] This invention provides an error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios. Please refer to [link / reference]. Figure 1 This includes the following steps:

[0079] S1. Collect the original radar frame set output by the vehicle-mounted millimeter-wave radar, establish a joint time axis, perform coordinate registration, geometric projection and noise removal on the radar data, and generate the original point cloud dataset Pcl.

[0080] S2. Using the original point cloud dataset Pcl as input, establish a candidate target set Can. Obtain a large target set Big through volume, scale and stability conditions, and generate an occlusion region set Occ. Form a candidate scattering point set Csp within the occlusion region set Occ. Filter the weak scattering point set Sca according to signal strength and stability, and establish a boundary correspondence set Bdr with the large target set Big.

[0081] S3. Based on the weak scattering point set Sca, group the trajectory candidate set Trc according to spatial and temporal conditions; perform fitting calculations on the trajectory candidate set Trc to obtain velocity and acceleration estimates, and obtain the trajectory fitting set Fit through residual error screening.

[0082] S4. Calculate the compensation amount using the time synchronization error Ets and the calibration residual Ecl to form the trajectory compensation amount set Com, and apply it to the trajectory fitting set Fit to obtain the corrected trajectory set Cor;

[0083] S5. The correction trajectory set Cor is used as the final trajectory result for error correction and is directly called in scene reconstruction and target recognition.

[0084] In this embodiment, steps S1 to S5 achieve accurate error correction of the vehicle-mounted millimeter-wave radar output in complex traffic scenarios. Specifically, a candidate target set Can is established using the original point cloud dataset Pcl, and an occlusion region set Occ is generated. This ensures that even when large targets such as buses and trucks create occlusion, a stable weak scattering point set Sca can still be extracted from the candidate scattering point set Csp. Furthermore, the boundary correspondence set Bdr ensures that weak traces within the occlusion region are not ignored, thus avoiding the problem of complete target loss in tunnel entrances or scenarios where large vehicles converge. Further, this method maintains the continuity and reliability of the trajectory in both time series and spatial coordinates by establishing a trajectory candidate set Trc and a trajectory fitting set Fit. The trajectory compensation set Com, generated by combining the time synchronization error Ets and the calibration residual Ecl, is applied to the trajectory fitting set Fit. The resulting corrected trajectory set Cor effectively eliminates drift and misalignment caused by radar sampling delay and installation angle deviation. Finally, the corrected trajectory set Cor is used as the output for scene reconstruction and target recognition, so that vehicles can still obtain temporally aligned and spatially accurate error-corrected final trajectory results in complex traffic environments such as multi-lane highways at night and intersections in rainy or foggy weather, ensuring that the detection of surrounding vehicles and obstacles is not delayed or deviated.

[0085] Example 2

[0086] Specifically: S1 includes S11;

[0087] S11. During vehicle operation, the vehicle-mounted millimeter-wave radar collects signals according to a set sampling period, and after receiving, demodulating and initially formatting, outputs the radar raw frame set Rfr.

[0088] The system clock generates a timestamp sequence Tst and establishes a joint time axis Tim, so that the radar's original frame set Rfr and the original motion data sequence Mrd are consistent in the time dimension.

[0089] The alignment difference of the data in the original radar frame set Rfr under the timestamp sequence Tst is calculated to obtain the synchronization offset Dlt, and the synchronization offset Dlt is used to map the data to the joint time axis Tim for timestamp binding.

[0090] S1 further includes S12;

[0091] S12. Load the external and internal parameters saved during vehicle loading to form coordinate system calibration parameters Cal. Use rigid body coordinate transformation to register the millimeter-wave radar coordinate system with the vehicle body coordinate system to ensure coordinate uniformity.

[0092] Geometric projection is performed on the original radar frame set Rfr based on the coordinate projection mapping method, and noise removal is performed on isolated points and outliers in the point cloud data based on the statistical outlier removal method to generate the original point cloud dataset Pcl, which is used to keep consistent with the vehicle body coordinates.

[0093] Based on the mean square error calculation method, the time synchronization error Ets is obtained by performing difference processing on the joint time axis Tim and the synchronization offset Dlt.

[0094] The original radar frame set Rfr is fitted with the coordinate system calibration parameter Cal using the least squares fitting method to obtain the calibration residual Ecl.

[0095] The joint time axis Tim is statistically processed using a frame coverage statistical method to obtain the data coverage Cov.

[0096] In this embodiment, the integrity and consistency of radar information are ensured at the data source level through steps S11 and S12. A joint time axis Tim is established using the original radar frame set Rfr and the timestamp sequence Tst generated by the system clock. Through correction of the synchronization offset Dlt, strict alignment of Rfr and the original motion data sequence Mrd in the time dimension is achieved. This allows different sensors to respond synchronously to instantaneous changes in traffic scenarios when used together, such as when vehicles merge into lanes at speeds exceeding 100 km / h on highways, avoiding misjudgments of target positions due to time drift. Furthermore, the coordinate system calibration parameters Cal saved during vehicle installation are loaded, and the original point cloud dataset Pcl, generated through geometric projection and noise removal methods, is used to ensure that the point cloud results remain consistent with the vehicle's coordinate system, eliminating interference from isolated and outlier points. This ensures a stable point cloud structure can still be obtained even in complex scenarios with strong metal reflections under urban overpasses. Meanwhile, the time synchronization error Ets calculated based on the joint time axis Tim, the calibration residual Ecl calculated based on the fitting, and the data coverage Cov obtained by statistics provide quantitative indicators for subsequent trajectory compensation and reliability analysis, enabling vehicles to evaluate data quality in real time during long-term operation, thereby maintaining a continuous and stable perception of the traffic environment in applications that span day and night and weather conditions.

[0097] Example 3

[0098] Specifically: S2 includes S21;

[0099] S21. Using the original point cloud dataset Pcl as input, perform density aggregation and time series association on the spatial points to obtain multiple point cloud clusters, and combine these point cloud clusters into a candidate target set Can.

[0100] Within the candidate target set Can, calculate the volume parameter Vol, aspect ratio Rat, and temporal stability Sta for each point cloud cluster;

[0101] The volume parameter Vol is obtained by multiplying the length, width, and height of the three-dimensional bounding box of the point cloud cluster. The calculation form is Vol = length × width × height.

[0102] The aspect ratio Rat is obtained by the ratio of the maximum side length to the minimum side length of the point cloud cluster bounding box, and is calculated as Rat = maximum side length / minimum side length.

[0103] The temporal stability Sta is obtained by statistically analyzing the ratio of the number of times a point cloud cluster appears in consecutive frames to the total number of frames. The calculation form is Sta = number of occurrence frames ÷ total number of frames.

[0104] When the following three conditions are met:

[0105] Volume parameter Vol > volume threshold Vth;

[0106] The aspect ratio Rat falls within the aspect ratio threshold range Rth_min≤Rat≤Rth_max;

[0107] Time stability Sta ≥ stability threshold Sth;

[0108] The point cloud clusters are labeled and then incorporated into the large target set Big.

[0109] Based on the geometric outline of the large target set Big, projection calculations are performed in the vehicle coordinate system to calculate the occluded spatial region, forming the occluded region set Occ, and maintaining the correspondence with the large target set Big.

[0110] Among them, the volume parameter Vol of all point cloud clusters in the candidate target set Can is statistically analyzed, and its mean value is multiplied by a fixed coefficient K. Wherein (K>1, for example K=2), the calculation formula is Vth=average Vol×K, which is used to distinguish large targets from ordinary vehicles.

[0111] Rth_min and Rth_max represent the minimum and maximum aspect ratio thresholds, respectively, which are determined by the geometric features of common large vehicles, including buses and trains, and are used to limit the proportion of point cloud clusters. The specific values ​​are set by the user.

[0112] The stability threshold Sth is set empirically based on actual application scenarios. For example, setting Sth=0.7 means that a target is considered a stable large target only if it remains present in at least 70% of consecutive frames with a stability threshold Sth=70%.

[0113] S2 further includes S22;

[0114] S22. Using the occlusion region set Occ and the original point cloud dataset Pcl as input, extract all points that fall within the occlusion region set Occ and combine them to form a candidate scattering point set Csp.

[0115] Within the candidate scattering point set Csp, two parameters are calculated for each point: signal strength Sig and frame occurrence count Fre;

[0116] Among them, the signal strength Sig is determined by the amplitude of the millimeter-wave radar echo signal. The calculation form is Sig = echo energy value ÷ maximum energy normalization. This parameter is used to measure the scattering intensity of the spot.

[0117] Frame occurrence count Fre: This is obtained by counting the number of times the point appears in consecutive frames. The calculation form is Fre = number of occurrence frames ÷ total number of frames. It is used to measure the stability of the point over time.

[0118] Define filtering conditions for the candidate scattering point set Csp, and filter each point trace according to the filtering conditions:

[0119] Filtering condition 1: Signal strength Sig < strength threshold Sth;

[0120] Filtering condition 2: Frame occurrence count Fre ≥ stability threshold Fth;

[0121] Among them, the intensity threshold Sth is used to divide the energy threshold between strong and weak scattering points, and the specific value is set by the user; the stability threshold Fth is used to determine whether weak scattering points remain in continuous frames. For example, setting Fth=0.6 means that the point remains in at least 60% of the continuous frames, and the specific value is set by the user.

[0122] When screening conditions one and two are met, the point is marked and included in the established weak scattering point set Sca;

[0123] The three-dimensional coordinates of each point in the weak scattering point set Sca are compared with the geometric outline of the large target set Big.

[0124] When the spatial location of the point trace in the weak scattering point set Sca falls within the boundary extension range of the large target set Big, a mapping relationship between the point trace and the corresponding large target set Big is established in the boundary corresponding set Bdr to represent the correspondence between the weak scattering point and the boundary of the occluder.

[0125] It should be noted that:

[0126] Big target set: The set of occlusion objects such as buses and trucks obtained from the candidate target set Can by determining their volume, shape ratio and time stability.

[0127] Boundary correspondence set Bdr: A set established through geometric comparison, used to describe the spatial correspondence between the set of weak scattering points Sca and the boundary of the set of large targets Big;

[0128] By performing point-by-point calculation and threshold screening of the candidate scattering point set Csp in step S22, unstable noise points can be effectively eliminated, while stable weak scattering points can be retained. Furthermore, a correspondence can be formed with the large target set Big through boundary mapping, so that the weak information in the occluded scene can be utilized, providing reliable input data for subsequent trajectory inference and error correction.

[0129] In this embodiment, through steps S21 and S22, the dual utilization of large targets and weak scattering points is achieved in complex occluded traffic scenarios. First, based on the volume parameter Vol, aspect ratio Rat, and time stability Sta of the original point cloud dataset Pcl, clusters that meet the conditions in the candidate target set Can are extracted into a large target set Big, and further, an occlusion region set Occ is generated. This allows for the clear identification of the spatial extent of occluded objects in environments where vehicles are traveling through tunnel entrances or bus stops with dense large occlusions. Second, the candidate scattering point set Csp is obtained by calculating the association between the occlusion region set Occ and Pcl. Through joint filtering using signal strength Sig and frame occurrence frequency Fre, a weak scattering point set Sca is established, and a boundary correspondence set Bdr is established with the large target set Big. This allows targets that are originally difficult to observe directly to obtain boundary information compensation through weak scattering points even under strong occlusion conditions. For example, when a truck completely obscures a small motorcycle in an adjacent lane, this method can utilize the spatial correspondence between Sca and Big to recover the motorcycle's motion signature in the boundary correspondence set Bdr, providing additional observable evidence for subsequent trajectory inference. Thus, step S2 not only effectively distinguishes between large occluded targets and ordinary vehicles, but also overcomes the limitation of "occlusion equals loss" in traditional methods through the stable utilization of weak scattering points, enabling the system to possess stronger occlusion robustness and target reconstruction capabilities in complex traffic scenarios.

[0130] Example 4

[0131] Specifically: S3 includes S31;

[0132] S31. Using the set of weak scattering points Sca as input, perform joint calculations on the points in the set in three-dimensional coordinates and time series to obtain the spatial distance Dis and time interval Int.

[0133] The spatial distance Dis represents the three-dimensional Euclidean distance between adjacent points; the time interval Int represents the sampling time difference between adjacent points.

[0134] When the spatial distance Dis ≤ the preset distance threshold Dth and the time interval Int ≤ the preset time threshold Tth, the points are grouped into the same trajectory cluster;

[0135] Group all points that meet the conditions to form a candidate trajectory set Trc;

[0136] The dots are grouped as follows:

[0137] Select an ungrouped point from the weak scattering point set Sca as the initial trajectory point;

[0138] Then, search the weak scattering point set Sca for all points that satisfy the spatial distance Dis ≤ distance threshold Dth and the time interval Int ≤ time threshold Tth, and add these points to the same group;

[0139] Repeat the search steps for newly added points in the group until no more points meet the conditions.

[0140] The set of points is marked as a trajectory cluster and removed from the weak scattering point set Sca;

[0141] Repeat the above process until all traces in the weak scattering point set Sca are grouped, resulting in multiple trajectory clusters, which are combined to form the trajectory candidate set Trc.

[0142] It should be noted that:

[0143] Trajectory cluster: A group of points consisting of a series of spatially adjacent and temporally continuous weak scattering points;

[0144] Trajectory candidate set Trc: A set composed of multiple trajectory clusters, which serves as the input for subsequent trajectory fitting;

[0145] This step organizes the scattered points in the weak scattering point set Sca into a trajectory cluster with spatiotemporal continuity through an explicit iterative grouping mechanism, thus providing an operational data foundation for subsequent fitting calculations.

[0146] S3 further includes S32;

[0147] S32. Using the trajectory candidate set Trc as input, perform trajectory fitting analysis on each trajectory cluster in the time series dimension to obtain velocity estimate Vel and acceleration estimate Acc.

[0148] The velocity estimate Vel is calculated from the spatial displacement and time interval of adjacent points, and the calculation form is Vel = displacement ÷ time interval;

[0149] The acceleration estimate Acc is calculated from the changes in adjacent velocities and the time interval, and the calculation form is Acc = (Vel2 - Vel1) ÷ time interval; Vel2 and Vel1 represent the velocities of point 2 and point 1, respectively.

[0150] The trajectory curves of the time series positions of the trajectory clusters are calculated based on the least squares fitting method, the fitting results are obtained, and the residual error value Err of each trajectory is recorded.

[0151] The obtained residual error value Err is then compared and filtered with the preset fitting error threshold Eth to obtain the filtered trajectory fitting set Fit;

[0152] The comparison and screening process was conducted in the following ways:

[0153] When the residual error Err ≤ the fitting error threshold Eth, the residual error value Err of the trajectory cluster is recorded in the trajectory fitting set Fit;

[0154] If the residual error Err is greater than the fitting error threshold Eth, then the trajectory cluster is discarded and not included in the trajectory fitting set Fit;

[0155] It should be noted that:

[0156] The residual error Err is the deviation between the fitted curve and the actual points, used to evaluate the fitting accuracy.

[0157] The fitting error threshold Eth is the upper limit used to determine whether the fitted trajectory is valid.

[0158] This step transforms scattered candidate trajectory clusters into continuously fitted trajectories, and retains reliable trajectories through residual error screening, thereby ensuring the reliability and traceability of input data for subsequent error correction.

[0159] In this embodiment, the continuity and precise selection of weak scattering point trajectories are achieved through the combined processing of steps S31 and S32. First, based on the three-dimensional coordinates and time series calculation of the weak scattering point set Sca, scattered points are aggregated according to the dual conditions of spatial distance Dis and time interval Int, forming multiple trajectory clusters with spatiotemporal continuity. These clusters are combined into a trajectory candidate set Trc, thus organizing the originally irregular point distribution into an analyzable trajectory sequence. Subsequently, the dynamic characteristics of the trajectory candidate set Trc are calculated using velocity estimation Vel and acceleration estimation Acc, and a continuous trajectory curve is obtained through the least squares fitting method. Simultaneously, a high-confidence trajectory fitting set Fit is generated by comparing the residual error value Err with the fitting error threshold Eth. This process not only makes the trajectory continuous and interpretable in time and space, but also effectively eliminates abnormal trajectories through residual constraints, avoiding interference from noisy point clouds on subsequent error correction. For example, in rainy weather, raindrops may form random scattering points. If these points are directly involved in trajectory analysis, it will lead to incorrect estimation of the target velocity. However, through the above fitting and residual screening, raindrop trajectories are eliminated due to excessive residual error values ​​Err. The final retained trajectory fitting set Fit can provide stable and reliable data input for subsequent compensation calculations.

[0160] Example 5

[0161] Specifically: S4 includes S41;

[0162] S41. Convert the time synchronization error Ets into a time compensation value Tco, and match it with each trajectory in the trajectory fitting set Fit, in order to correct the uniform offset of the trajectory on the time axis.

[0163] The calibration residual Ecl is converted into a calibration compensation value Cco and matched with each trajectory in the trajectory fitting set Fit, which is used to correct the trajectory offset in spatial coordinates.

[0164] Establish a correspondence between the time compensation value Tco and the calibration compensation value Cco and each trajectory in the trajectory fitting set Fit, forming the trajectory compensation quantity set Com;

[0165] It should be noted that:

[0166] The time synchronization error Ets represents the fixed deviation between the radar sampling clock and the vehicle system's global clock. To correct this deviation, a compensation value needs to be defined: the time compensation value Tco. This compensation value is directly given by the time synchronization error Ets, that is, the time compensation value Tco = the time synchronization error Ets. Its function is to uniformly adjust all trajectory timestamps so that the trajectory time axis is aligned with the global clock.

[0167] The calibration residual Ecl represents the fixed offset between the radar installation position or attitude and the vehicle reference coordinate system. In order to eliminate this offset, a compensation value needs to be defined: the calibration compensation value Cco. This compensation value is directly given by the calibration residual Ecl, that is, the calibration compensation value Cco = calibration residual Ecl. Its function is to translate or rotate the trajectory space coordinates as a whole to make them consistent with the vehicle reference coordinate system.

[0168] S4 also includes S42;

[0169] S42. Apply a time compensation value Tco to all trajectories in the trajectory fitting set Fit to obtain the time-aligned trajectory results.

[0170] A uniform calibration compensation value Cco is applied to all trajectories in the trajectory fitting set Fit to obtain the spatially corrected trajectory results, which are defined as the corrected trajectory set Cor.

[0171] It should be noted that:

[0172] The correction trajectory set Cor is the final trajectory set obtained by processing the time compensation value Tco and the calibration compensation value Cco in the trajectory compensation amount set Com, based on the trajectory fitting set Fit.

[0173] This step enables unified correction of the trajectory in both the time and spatial domains, ensuring that trajectory results in complex traffic scenarios are no longer limited by sensor delays or installation deviations, and providing highly reliable data support for vehicle perception and decision-making.

[0174] S5 includes S51;

[0175] S51. Using the correction trajectory set Cor as input, it serves as the final trajectory result for subsequent use by the vehicle system. During scene reconstruction, the correction trajectory set Cor is directly called, which can avoid the accumulation of errors caused by time drift and spatial deviation. At the same time, the correction trajectory set Cor is provided to the target detection and tracking stage to represent the position and motion state of dynamic targets, thereby providing high-precision basic data support for path prediction and obstacle avoidance control.

[0176] In this embodiment, through the coordinated processing of steps S41, S42, and S51, dual compensation and final correction of the trajectory in both the time and spatial domains are achieved. Specifically, the time synchronization error Ets is converted into a time compensation value Tco and applied to all trajectories, thereby ensuring that the trajectory timestamp is strictly consistent with the global clock and avoiding overall trajectory drift caused by radar sampling delay. Simultaneously, the calibration residual Ecl is converted into a calibration compensation value Cco and applied to the trajectory spatial coordinates, aligning the trajectory result with the vehicle reference coordinate system and eliminating spatial offset caused by sensor installation position or attitude. After the dual compensation processing, the resulting corrected trajectory set Cor is directly transmitted to the vehicle perception and control system for scene reconstruction and target tracking, thereby ensuring that the description of the dynamic target position and motion state is no longer subject to the basic measurement error. For example, in complex scenarios with multiple parallel lanes in an urban environment, an uncorrected trajectory may cause the vehicle's position to lag behind the actual vehicle length by half a vehicle length due to radar delay, or cause targets in adjacent lanes to be misjudged as targets in the current lane due to installation deviation. However, by applying the corrected trajectory set Cor, the vehicle's actual lane position and trajectory can be accurately distinguished, enabling subsequent path prediction and obstacle avoidance control to be performed within a millisecond response time, avoiding decision-making errors caused by delays or deviations.

[0177] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for error correction of vehicle-mounted millimeter-wave radar in complex traffic scenarios, characterized in that: Includes the following steps: S1. Collect the original radar frame set output by the vehicle-mounted millimeter-wave radar, establish a joint time axis, perform coordinate registration, geometric projection and noise removal on the radar data, and generate the original point cloud dataset Pcl. S2. Using the original point cloud dataset Pcl as input, establish a candidate target set Can. Obtain a large target set Big through volume, scale and time stability conditions, and generate an occlusion region set Occ. Form a candidate scattering point set Csp within the occlusion region set Occ. Filter the weak scattering point set Sca according to signal strength and frame occurrence frequency, and establish a boundary correspondence set Bdr with the large target set Big. Based on the geometric outline of the large target set Big, projection calculations are performed in the vehicle coordinate system to calculate the occluded spatial region, forming the occluded region set Occ, and maintaining the correspondence with the large target set Big. The three-dimensional coordinates of each point in the weak scattering point set Sca are compared with the geometric outline of the large target set Big. When the spatial location of the point trace in the weak scattering point set Sca falls within the boundary extension range of the large target set Big, a mapping relationship between the point trace and the corresponding large target set Big is established in the boundary corresponding set Bdr. S3. Based on the weak scattering point set Sca, group the trajectory candidate set Trc according to spatial and temporal conditions; perform fitting calculations on the trajectory candidate set Trc to obtain velocity and acceleration estimates, and obtain the trajectory fitting set Fit through residual error screening. S4. Calculate the compensation amount using the time synchronization error Ets and the calibration residual Ecl to form the trajectory compensation amount set Com, and apply it to the trajectory fitting set Fit to obtain the corrected trajectory set Cor; S5. The correction trajectory set Cor is used as the final trajectory result for error correction and is directly called in scene reconstruction and target recognition.

2. The error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 1, characterized in that: Step S1 includes: S11. During vehicle operation, the vehicle-mounted millimeter-wave radar collects signals according to a set sampling period, and after receiving, demodulating and initially formatting, outputs the radar raw frame set Rfr. The system clock generates a timestamp sequence Tst and establishes a joint time axis Tim, so that the radar's original frame set Rfr and the original motion data sequence Mrd are consistent in the time dimension. The alignment difference of the data in the original radar frame set Rfr under the timestamp sequence Tst is calculated to obtain the synchronization offset Dlt, and the synchronization offset Dlt is used to map the data to the joint time axis Tim for timestamp binding.

3. The error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 2, characterized in that: Step S1 further includes: S12. Load the external and internal parameters saved during vehicle installation to form coordinate system calibration parameters Cal, and register the millimeter-wave radar coordinate system with the vehicle body coordinate system through rigid body coordinate transformation. Geometric projection is performed on the original radar frame set Rfr based on the coordinate projection mapping method, and noise removal is performed on isolated points and outliers in the point cloud data based on the statistical outlier removal method to generate the original point cloud dataset Pcl. Based on the mean square error calculation method, the time synchronization error Ets is obtained by performing difference processing on the joint time axis Tim and the synchronization offset Dlt. The original radar frame set Rfr is fitted with the coordinate system calibration parameter Cal using the least squares fitting method to obtain the calibration residual Ecl. The joint time axis Tim is statistically processed using a frame coverage statistical method to obtain the data coverage Cov.

4. The method for error correction of vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 3, characterized in that: Step S2 includes: S21. Using the original point cloud dataset Pcl as input, perform density aggregation and time series association on the spatial points to obtain multiple point cloud clusters, and combine these point cloud clusters into a candidate target set Can. Within the candidate target set Can, calculate the volume parameter Vol, aspect ratio Rat, and temporal stability Sta for each point cloud cluster; The volume parameter Vol is obtained by multiplying the length, width, and height of the three-dimensional bounding box of the point cloud cluster. The calculation form is Vol = length × width × height. The aspect ratio Rat is obtained by the ratio of the maximum side length to the minimum side length of the point cloud cluster bounding box, and is calculated as Rat = maximum side length / minimum side length. The temporal stability Sta is obtained by statistically analyzing the ratio of the number of times a point cloud cluster appears in consecutive frames to the total number of frames. The calculation form is Sta = number of occurrence frames ÷ total number of frames. When the following three conditions are met: Volume parameter Vol > volume threshold Vth; The aspect ratio Rat falls within the aspect ratio threshold range Rth_min≤Rat≤Rth_max; Time stability Sta ≥ time stability threshold Sth; The point cloud clusters are labeled and then incorporated into the large target set Big. The volume threshold Vth is obtained by statistically analyzing the volume parameters Vol of all point cloud clusters in the candidate target set Can, taking their average value and multiplying it by a fixed coefficient K, where the fixed coefficient K > 1. Rth_min and Rth_max represent the minimum and maximum aspect ratio thresholds, respectively, and are determined by the geometric shape features of large vehicles, including buses and trains, to limit the proportion of point cloud clusters. The specific values ​​are set by the user. The time stability threshold Sth is set empirically in real-world application scenarios, indicating that a target is considered a stable large target only if it remains present for at least consecutive frames within the time stability threshold Sth.

5. The error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 4, characterized in that: Step S2 further includes: S22. Using the occlusion region set Occ and the original point cloud dataset Pcl as input, extract all points that fall within the occlusion region set Occ and combine them to form a candidate scattering point set Csp. Within the candidate scattering point set Csp, two parameters are calculated for each point: signal strength Sig and frame occurrence count Fre; Among them, the signal strength Sig is determined by the amplitude of the millimeter-wave radar echo signal; Frame occurrence count Fre: Obtained by counting the number of times the point appears in consecutive frames; Define filtering conditions for the candidate scattering point set Csp, and filter each point trace according to the filtering conditions: Filtering condition 1: Signal strength Sig < strength threshold Ith; Filtering condition 2: Frame occurrence count Fre ≥ frame occurrence stability threshold Fth; Among them, the intensity threshold Ith is the energy threshold used to distinguish between strong and weak scattering points, and the specific value is set by the user; the frame occurrence stability threshold Fth is used to determine whether weak scattering points remain in consecutive frames, and the specific value is set by the user. When screening conditions one and two are met, the points are marked and included in the established weak scattering point set Sca.

6. The method for error correction of vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 5, characterized in that: Step S3 includes: S31. Using the set of weak scattering points Sca as input, perform joint calculations on the points in the set in three-dimensional coordinates and time series to obtain the spatial distance Dis and time interval Int. The spatial distance Dis represents the three-dimensional Euclidean distance between adjacent points; the time interval Int represents the sampling time difference between adjacent points. When the spatial distance Dis ≤ the preset distance threshold Dth and the time interval Int ≤ the preset time threshold Tth, the points are grouped into the same trajectory cluster; Group all points that meet the conditions to form a candidate trajectory set Trc.

7. The method for error correction of vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 6, characterized in that: Step S3 further includes: S32. Using the trajectory candidate set Trc as input, perform trajectory fitting analysis on each trajectory cluster in the time series dimension to obtain velocity estimate Vel and acceleration estimate Acc. The velocity estimate Vel is calculated from the spatial displacement and time interval between adjacent points; The acceleration estimate Acc is calculated from the changes in adjacent velocities and the time interval. The trajectory curves of the time series positions of the trajectory clusters are calculated based on the least squares fitting method, the fitting results are obtained, and the residual error value Err of each trajectory is recorded. The obtained residual error value Err is then compared and filtered with the preset fitting error threshold Eth to obtain the filtered trajectory fitting set Fit; The comparison and screening process was conducted in the following ways: When the residual error Err ≤ the fitting error threshold Eth, the residual error value Err of the trajectory cluster is recorded in the trajectory fitting set Fit; If the residual error Err is greater than the fitting error threshold Eth, then the trajectory cluster is discarded and not included in the trajectory fitting set Fit.

8. The method for error correction of vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 7, characterized in that: Step S4 includes: S41. Convert the time synchronization error Ets into a time compensation value Tco, and match it with each trajectory in the trajectory fitting set Fit, in order to correct the uniform offset of the trajectory on the time axis. The calibration residual Ecl is converted into a calibration compensation value Cco and matched with each trajectory in the trajectory fitting set Fit, which is used to correct the trajectory offset in spatial coordinates. Establish a correspondence between the time compensation value Tco, the calibration compensation value Cco, and each trajectory in the trajectory fitting set Fit, forming the trajectory compensation quantity set Com.

9. The error correction method for vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 8, characterized in that: Step S4 further includes: S42. Apply a time compensation value Tco to all trajectories in the trajectory fitting set Fit to obtain the time-aligned trajectory results. A calibration compensation value Cco is uniformly applied to all trajectories in the trajectory fitting set Fit to obtain the spatially corrected trajectory results, which are defined as the corrected trajectory set Cor.

10. The method for error correction of vehicle-mounted millimeter-wave radar in complex traffic scenarios according to claim 9, characterized in that: Step S5 includes: S51. Taking the correction trajectory set Cor as input, it is used as the final trajectory result for subsequent use by the vehicle system. During the scene reconstruction process, the correction trajectory set Cor is directly called. At the same time, the correction trajectory set Cor is provided to the target detection and tracking stage to represent the position and motion state of the dynamic target.

Images