Wide-area vehicle track splicing method and system based on millimeter-wave radar group

Abstract

The invention belongs to the technical field of intelligent traffic, discloses a wide-area vehicle track splicing method based on a millimeter-wave radar group, aims to solve the problems that the detection range of single equipment in a road wide-area range is limited and shielding exists in a traditional visual splicing method, overcomes the limitation of the prior art in a complex traffic environment, and improves the vehicle track splicing efficiency. And moreover, the problem of cross-equipment track splicing of the millimeter-wave radar is solved, and a data source is provided for further analysis and detection by using the millimeter-wave radar. The method comprises the following steps: acquiring motion information including relative position, speed and acceleration and a unique identifier through vehicle data in a millimeter wave radar detection range; and the detection areas are divided into different categories, and a set to which the tracks belong is judged, so that targeted and specific matching and splicing can be carried out. Therefore, more continuous and wide-coverage vehicle track data is realized, and more reliable data support is provided for operation of an intelligent traffic system.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent transportation, and particularly relates to a wide-area vehicle trajectory stitching method and system based on a millimeter-wave radar group. Background Art

[0002] At present, the construction of road intelligence has become one of the hot topics in the transportation industry, and this process is inseparable from effective sensing technologies. As a core component of the intelligent transportation system, the perception of vehicle position and state is crucial. Real-time and accurate position information can not only greatly improve the efficiency and accuracy of road operation monitoring, but also provide important support for command and dispatch, congestion detection, traffic accident warning, etc., ultimately helping to optimize the operation of the road network and improve the intelligent level of the transportation system.

[0003] In recent years, video surveillance technology has made remarkable progress, and the application of cameras in the transportation field has gradually increased and is widely used for road monitoring and vehicle identification. However, video sensing technology also faces many challenges. First of all, high-definition cameras are usually expensive and are easily affected by external factors such as weather changes and lighting conditions, which may lead to blurred images or data loss, affecting subsequent detection and recognition accuracy. Secondly, the use of a large number of ordinary cameras, if not effectively integrated and optimized with other sensing technologies, may cause waste of resources and low system efficiency. Compared with video technology, millimeter-wave radar technology shows unique advantages in this field. Millimeter-wave radar can effectively penetrate the influence of bad weather conditions (such as rain and snow) and low-light environments and has strong anti-interference ability. In addition, millimeter-wave radar can work in coordination with various sensing means such as existing video surveillance systems and ground sensing devices to further improve the sensing effect and the robustness of the overall system.

[0004] At present, the maximum detection distance of a single millimeter-wave radar is about 250 meters. By deploying multiple millimeter-wave radars and performing trajectory stitching, all-weather and full-coverage vehicle position perception can be achieved. This technical solution shows excellent application prospects in many actual scenarios. However, there are still some deficiencies in the existing trajectory stitching algorithms. The current stitching methods usually assume good road conditions, low traffic flow, and no interference factors such as obstacles, and are applicable to relatively simple environments. However, when facing complex situations such as high traffic flow, obstacles, and curves, the performance of the existing algorithms will decrease significantly, making it difficult to meet the requirements of tracking targets with high precision throughout the whole process and globally. Therefore, how to improve the robustness and accuracy of the stitching algorithm under complex road conditions remains an urgent technical problem to be solved.

[0005] Through the above analysis, the problems and defects existing in the prior art are as follows:

[0006] When facing complex situations such as large - volume traffic, obstacles, and curves, the performance of existing algorithms will significantly decline, making it difficult to meet the requirements of tracking targets with high precision throughout the whole process and globally. Summary of the Invention

[0007] In view of the problems existing in the prior art, the present invention provides a wide - area vehicle trajectory stitching method based on a millimeter - wave radar group.

[0008] The present invention is implemented as follows. A wide - area vehicle trajectory stitching method based on a millimeter - wave radar group includes:

[0009] Step 1: Obtain the trajectory information of vehicles through a preset millimeter - wave radar group on the road;

[0010] Specifically, the millimeter - wave radar detects targets in its front fan - shaped area, and obtains various motion characteristics of the target vehicle through reflection, including the id of the detected target as its unique identifier, the lateral offset and longitudinal offset of the target relative to the current millimeter - wave radar, the relative speed and acceleration of the current motion state of the target, and timestamp information;

[0011] Step 2: Classify vehicle trajectories within a wide area;

[0012] Specifically, for vehicle trajectories within the range, identify vehicles that suddenly disappear, suddenly appear, enter the detection area, enter the non - overlapping area, enter the overlapping area, and leave the detection area;

[0013] Step 3: Determine the area and set to which the vehicle trajectory belongs;

[0014] Specifically, divide the millimeter - wave radar detection area into the initial entry into the radar detection area, non - overlapping area, front overlapping area, rear overlapping area, and leaving the radar detection area; in order to enable targeted stitching of millimeter - wave radar trajectory data in various scenarios, divide the trajectories obtained by the millimeter - wave radar using different sets, including the sudden disappearance set SDC, sudden appearance set SAC, overlapping area set LC, and global tracking set GTC;

[0015] Step 4: Perform multi - point matching in the overlapping area;

[0016] Specifically, perform multi - point matching in the overlapping area, and update the vehicle trajectory information that matches successfully in the corresponding set in real - time; the overlapping area matching is only carried out between adjacent millimeter - wave radars. For example, in the i - th overlapping area, the relevant overlapping set is LC i,front 、LC i,back ;Each recorded trajectory in LC i,front is compared with LC i,backPerform matching by setting multiple conditions and making judgments according to the matching rules. After completion of the matching, change the information in the corresponding set. The set change rules are as follows: a. If LC i,front matches successfully, remove the corresponding vehicle record in the suddenly disappearing set SDC; b. If LC i,front and LC i,back both match successfully, modify the trajectory concatenated with front in the global tracking set GTC to the information corresponding to back; After the matching is completed, add the vehicles that have not been successfully matched to the suddenly appearing set SAC according to the following rule (formula);

[0017] Step 5, Concatenate the target trajectories in the non-overlapping area;

[0018] Specifically, it includes: concatenating the vehicle trajectories in the non-overlapping area, which also includes the trajectory data that fails to match in the partially overlapping area. The appearance and disappearance of the trajectories may be separated by a relatively long distance due to certain reasons. First, use an improved linear trend model to fill in the trajectory data; on the other hand, due to the uncertainty of the corresponding relationship between the disappearing trajectory and the appearing trajectory, data filling can be performed between any disappearing trajectory and any appearing trajectory. In order to avoid the deviation between the predicted filled points and the previous historical trajectories, use the Kalman filter for processing to obtain more reliable trajectories; then, on this basis, use the relative distance and the length transformation value of the vehicle itself as weights, and use the extended Hungarian matching algorithm for concatenation.

[0019] Furthermore, the basis for determining the area to which the vehicle trajectory belongs is as follows:

[0020] For the first radar:

[0021]

[0022] For the last radar:

[0023]

[0024] For the remaining radars:

[0025]

[0026] In the above formula, inLap represents the area where the millimeter-wave radar is located; the values 1, 2, 3, 4, and 5 represent the rear overlapping area, non-overlapping area, front overlapping area, leaving the radar detection area, and initially entering the radar detection area respectively; x represents the coordinates of the target vehicle in the detection area; L1 and L2 represent the starting point and ending point of the detection area respectively; S1, S2, and S3 represent the detection area distance threshold, overlapping area distance threshold, and leaving detection area distance threshold respectively.

[0027] Furthermore, in the overlapping area, LCi,front , LC i,back The matching rules are as follows:

[0028] |point k,time,front -point j,time,back |≤Δtime

[0029] distance k,j ≤dist

[0030] |point k,speed,front -point j,speed,back |≤Δspeed

[0031] |point k,length,front -point j,length,back |≤Δlength

[0032] Judgment is made by setting conditions of time, distance, speed and length. In the formula, point k,time,front represents the timestamp corresponding to the k-th trajectory on a certain trajectory in the front overlapping area; time represents the timestamp corresponding to the j-th trajectory on a certain trajectory in the rear overlapping area; point j,time,back represents the threshold for determining that two timestamps are close enough; point k,speed,front represents the speed corresponding to the k-th trajectory on a certain trajectory in the front overlapping area; point j,speed,back represents the speed corresponding to the j-th trajectory on a certain trajectory in the rear overlapping area; speed represents the threshold for determining that two speed values are close enough; point k,length,front represents the vehicle body length collected by the k-th trajectory on a certain trajectory in the front overlapping area; point j,length,back represents the vehicle body length collected by the j-th trajectory on a certain trajectory in the rear overlapping area; Δlength represents the threshold for determining that two vehicle body lengths are close enough; distance k,j represents the distance between the k-th trajectory point and the j-th trajectory point; dist represents the threshold for determining that the distances between two vehicles are close enough.

[0033] Furthermore, the improved linear trend model is:

[0034] n=(trace begin,back -trace end,front ) / interval

[0035]

[0036] point t+n =pa t +npb t

[0037] Among them, n is the number of data to be filled between the disappearance and appearance of the trajectory; trace begin,back and trace end,back respectively represent the start time of the appearing trajectory and the end time of the disappearing trajectory; interval represents the time interval between data; is the moving average of the current time t; point i is the trajectory point at time i; is the second moving average; pa t and pb t are smoothing coefficients.

[0038] Furthermore, the relative distance between the trajectories is calculated through the last point of the trajectory obtained by filling and performing Kalman filtering, that is, pointNew t+n and the distance is in the Mars coordinate system.

[0039] Furthermore, the transformation rule of the extended Hungarian matching algorithm is expressed as:

[0040] maxDist = MAX(dist i,j ) + aMIN(dist i,j )

[0041] maxDiff = MAX(|lenDiff i,j |) + bMIN(lenDiff i,j |)

[0042]

[0043] In the formula, maxDist represents the maximum distance between the disappearance set and the appearance set plus the offset; maxDiff represents the maximum vehicle length difference between the disappearance set and the appearance set plus the offset; weight i,j represents the matching weight between the disappearance set and the appearance set; where the parameters a and b are both offset parameters, c is the weight parameter of the distance, and e is the weight parameter of the vehicle length difference.

[0044] Another object of the present invention is to provide a wide-area vehicle trajectory stitching system based on a millimeter-wave radar group, including:

[0045] A trajectory information acquisition module for acquiring vehicle trajectory information through a preset millimeter-wave radar group on the road;

[0046] A classification module for classifying vehicle trajectories within a wide area;

[0047] A determination module for determining the area and set to which the vehicle trajectory belongs;

[0048] A matching module for performing multi-point matching in the overlapping area;

[0049] A splicing module for splicing the target trajectories in the non-overlapping area.

[0050] Another object of the present invention is to provide a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the method for wide-area vehicle trajectory splicing based on a millimeter-wave radar group.

[0051] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method for wide-area vehicle trajectory splicing based on a millimeter-wave radar group.

[0052] Another object of the present invention is to provide an information data processing terminal for implementing the wide-area vehicle trajectory splicing system based on a millimeter-wave radar group.

[0053] Combined with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0054] First, the present invention provides a method for wide-area vehicle trajectory splicing based on a millimeter-wave radar group. Several millimeter-wave radars are preset within the target road range. The information returned by each millimeter-wave radar is fully utilized, and the Hungarian matching algorithm is combined to splice the vehicle trajectories obtained by the millimeter-wave radars within the range, so as to obtain a more continuous and larger-range vehicle trajectory, providing a better data source for further research on vehicle trajectories in the field of intelligent transportation.

[0055] The present invention provides a vehicle trajectory stitching method based on a millimeter-wave radar group. This method aims to solve the problems of limited detection breadth of a single device in a wide road area and occlusion in traditional vision stitching methods, overcome the limitations of the prior art in complex traffic environments, and solve the problem of millimeter-wave radar cross-device trajectory stitching, providing a data source for further analysis and detection using millimeter-wave radar. Through vehicle data within the detection range of millimeter-wave radar, motion information including relative position, speed, and acceleration, as well as a unique identifier, are obtained; the detection area is divided into different categories and the set to which the trajectory belongs is determined for targeted specific matching and stitching. Multi-point matching is performed in the overlapping area to update the vehicle trajectory information of successfully matched vehicles in real time, and corresponding processing is carried out on the vehicle trajectories that are not successfully matched; in the non-overlapping area, the trajectory data is filled by a linear trend model and Kalman filtering, and an extended Hungarian matching algorithm is used for trajectory stitching to achieve more continuous and widely covered vehicle trajectory data, providing more reliable data support for the operation of the intelligent transportation system.

[0056] Second, the technical solution of the present invention realizes the acquisition of vehicle trajectories with a larger range and more continuity for wide-area roads by arranging multiple millimeter-wave radars within a certain range and collaborating for detection. Compared with the fragmentation problem faced by traditional single-radar data, this solution effectively improves the reliability and robustness of the trajectory stitching method, providing more powerful data support for detection and analysis in the field of intelligent transportation. By accurately tracking and stitching vehicle trajectories in complex road environments, potential safety risks can be reduced and the user experience can be improved, which is expected to further promote the technological progress and market expansion of related industries.

[0057] Aiming at the wide-area trajectory stitching of a millimeter-wave radar group, the present invention can still maintain a high stitching accuracy in complex scenarios such as high-traffic and continuous curves, breaking through the bottleneck of the existing mainstream technologies for global high-precision target tracking and stitching. In the past, in these complex situations, the detection accuracy of millimeter-wave radar or other sensing means (such as high-definition cameras) generally decreased, and it was difficult to obtain continuous and stable vehicle trajectory information. The stitching method proposed by the present invention significantly improves these deficiencies, providing a new technical direction and idea for subsequent related research.

[0058] Traditional road vehicle tracking and vehicle trajectory stitching technologies are often restricted by external interference factors in complex environments, such as changes in weather and lighting conditions. High-definition cameras are prone to image blurring or data loss in such scenarios, affecting the detection and recognition accuracy; while traditional stitching methods based on radar usually assume an ideal road surface environment and are difficult to cope with occlusions and complex terrains. The present invention can achieve effective stitching of vehicle trajectories even in complex road environments, high-traffic, and continuous curve scenarios through the collaborative work of multiple millimeter-wave radars, significantly improving the accuracy and robustness of trajectory stitching.

[0059] The present invention uses a collaborative detection method with millimeter-wave radars to achieve efficient vehicle trajectory stitching for a wide-area road environment, breaking the technical bias that traditionally relies solely on vision technology for road detection. This solution has high operability in terms of equipment selection and deployment, providing a new implementation path for intelligent transportation systems. By combining multi-source sensor information, the present invention not only improves the vehicle tracking accuracy but also provides richer data support for intelligent decision-making under complex road conditions, having important reference value for the further promotion and application of future intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flowchart of a wide-area vehicle trajectory stitching method based on a millimeter-wave radar group provided by an embodiment of the present invention.

[0061] Figure 2 is a block diagram of the structure of a wide-area vehicle trajectory stitching system based on a millimeter-wave radar group provided by an embodiment of the present invention.

[0062] Figure 3 is a complete operation flowchart provided by an embodiment of the present invention.

[0063] Figure 4 is a schematic diagram of dynamically determining the set to which a trajectory belongs provided by an embodiment of the present invention.

[0064] Figure 5 is a schematic diagram of filling in trajectories in a non-overlapping area provided by an embodiment of the present invention.

[0065] Figure 6 is a multi-target matching node diagram in a non-overlapping area provided by an embodiment of the present invention.

[0066] Figure 7 is a box plot of the results of vehicle trajectory stitching tests in an actual road scenario by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0068] As Figure 1 shown, a wide-area vehicle trajectory stitching method based on a millimeter-wave radar group provided by an embodiment of the present invention includes the following steps:

[0069] S101, obtaining trajectory information of vehicles through a preset millimeter-wave radar group on the road;

[0070] Specifically, the millimeter-wave radar detects targets within its front fan-shaped area, and obtains various motion characteristics of the target vehicle through reflection, including the id of the detected target as its unique identifier, the lateral offset and longitudinal offset of the target relative to the current millimeter-wave radar, the relative speed and acceleration of the current motion state of the target, and timestamp information;

[0071] S102, classify vehicle trajectories within a wide area;

[0072] Specifically, for vehicle trajectories within the range, identify vehicles that suddenly disappear, suddenly appear, enter the detection area, enter the non-overlap area, enter the overlap area, and leave the detection area;

[0073] S103, determine the area and set to which the vehicle trajectory belongs;

[0074] Specifically, divide the millimeter-wave radar detection area into the initial entry into the radar detection area, non-overlap area, front overlap area, rear overlap area, and leaving the radar detection area; in order to enable targeted stitching of millimeter-wave radar trajectory data in various scenarios, divide the trajectories obtained by the millimeter-wave radar into different sets, including the sudden disappearance set SDC, sudden appearance set SAC, overlap area set LC, and global tracking set GTC;

[0075] S104, perform multi-point matching within the overlap area;

[0076] Specifically, perform multi-point matching within the overlap area, and update the vehicle trajectory information that matches successfully in the corresponding set in real time; the overlap area matching is only performed between adjacent millimeter-wave radars. For example, in the i-th overlap area, the relevant overlap set is LC i,front 、LC i,back ; Each recorded trajectory in LC i,front is matched with LC i,back , and judged through the matching rules by setting multiple conditions; after the matching is completed, the information in the corresponding set is changed; the set change rules are as follows: a. If LC i,front matches successfully, remove the corresponding vehicle record in the sudden disappearance set SDC; b. If LC i,front and LC i,back both match successfully, modify the trajectory spliced by front to the information corresponding to back in the global tracking set GTC; after the matching is completed, add the vehicles that have not been successfully matched to the sudden appearance set SAC according to the following rules (formulas);

[0077] S105, splice the target trajectories in the non-overlap area;

[0078] Specifically, it includes: splicing vehicle trajectories in the non-overlapping area, which also includes trajectory data with failed matching in part of the overlapping area. The appearance and disappearance of trajectories may be separated by a relatively long distance due to certain reasons. First, an improved linear trend model is used to fill the trajectory data. On the other hand, due to the uncertainty of the correspondence between disappearing trajectories and emerging trajectories, data filling can be performed between any disappearing trajectory and any emerging trajectory. To avoid the deviation between the predicted filled points and the previous historical trajectories, Kalman filtering is used for processing to obtain more reliable trajectories. After that, on this basis, the relative distance and the length transformation value of the vehicle itself are used as weights, and the extended Hungarian matching algorithm is used for splicing.

[0079] The determination basis for the area to which the vehicle trajectory provided by the embodiment of the present invention belongs is as follows:

[0080] For the first radar:

[0081]

[0082] For the last radar:

[0083]

[0084] For the remaining radars:

[0085]

[0086] In the above formula, inLap represents the area where the millimeter-wave radar is located; the values 1, 2, 3, 4, and 5 respectively represent the rear overlapping area, non-overlapping area, front overlapping area, leaving the radar detection area, and initially entering the radar detection area; x represents the coordinates of the target vehicle in the detection area; L1 and L2 respectively represent the starting point and ending point of the detection area; S1, S2, and S3 respectively represent the detection area distance threshold, overlapping area distance threshold, and leaving detection area distance threshold.

[0087] The LC i,front , LC i,back matching rule provided by the embodiment of the present invention is:

[0088] |point k,time,front -point j,time,back |≤Δtime

[0089] distance k,j ≤dist

[0090] |point k,speed,front -point j,speed,back |≤Δspeed

[0091] |point k,length,front -pointj,length,back |≤Δlength

[0092] Judge by setting conditions of time, distance, speed and length, where point k,time,front represents the timestamp corresponding to the k-th trajectory on a certain trajectory in the front overlap area; time represents the timestamp corresponding to the j-th trajectory on a certain trajectory in the rear overlap area; point j,time,back represents the threshold for determining that two timestamps are close enough; point k,speed,front represents the speed corresponding to the k-th trajectory on a certain trajectory in the front overlap area; point j,speed,back represents the speed corresponding to the j-th trajectory on a certain trajectory in the rear overlap area; speed represents the threshold for determining that two speed values are close enough; point k,length,front represents the vehicle body length collected by the k-th trajectory on a certain trajectory in the front overlap area; point j,length,back represents the vehicle body length collected by the j-th trajectory on a certain trajectory in the rear overlap area; Δlength represents the threshold for determining that two vehicle body lengths are close enough; distance k,j represents the distance between the k-th trajectory point and the j-th trajectory point; dist represents the threshold for determining that the distances between two vehicles are close enough.

[0093] The improved linear trend model provided by the embodiment of the present invention is:

[0094] n=(trace begin,back -trace end,front ) / interval

[0095]

[0096] point t+n =pa t +npb t

[0097] where n is the number of data to be filled between the disappearance and appearance of the trajectory; trace begin,back , trace end,back respectively represent the start time of the appearance trajectory and the end time of the disappearance trajectory; interval represents the time interval between data; is the first moving average of the current time t; point i is the trajectory point at the i-th moment; is the second moving average; pa t , pb t are smoothing coefficients.

[0098] The relative distance between the trajectories provided by the embodiments of the present invention is calculated through the last point of the trajectory obtained by filling and performing Kalman filtering, that is, pointNew t+n The calculation is carried out, and the distance uses the Mars coordinate system.

[0099] The transformation rule of the extended Hungarian matching algorithm provided by the embodiments of the present invention is expressed as:

[0100] maxDist = MAX(dist i,j ) + aMIN(dist i,j )

[0101] maxDiff = MAX(|lenDiff i,j |) + bMIN(lenDiff i,j |)

[0102]

[0103] In the formula, maxDist represents the maximum distance plus the offset between the disappearance set and the appearance set; maxDiff represents the maximum vehicle length difference plus the offset between the disappearance set and the appearance set; weight i,j represents the matching weight between the disappearance set and the appearance set; where the parameters a and b are both offset parameters, c is the weight parameter of the distance, and e is the weight parameter of the vehicle length difference.

[0104] As Figure 2 shown, a wide-area vehicle trajectory stitching system based on a millimeter-wave radar group provided by the embodiments of the present invention includes:

[0105] A trajectory information acquisition module, configured to acquire vehicle trajectory information through a preset millimeter-wave radar group on the road;

[0106] A classification module, configured to classify vehicle trajectories within a wide area;

[0107] A determination module, configured to determine the area and set to which the vehicle trajectory belongs;

[0108] A matching module, configured to perform multi-point matching within the overlapping area;

[0109] A stitching module, configured to stitch target trajectories in the non-overlapping area.

[0110] Another object of the present invention is to provide a computer device, where the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the wide-area vehicle trajectory stitching method based on a millimeter-wave radar group.

[0111] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to execute the steps of the method for wide-area vehicle trajectory stitching based on a millimeter-wave radar group.

[0112] Another object of the present invention is to provide an information data processing terminal for implementing the wide-area vehicle trajectory stitching system based on a millimeter-wave radar group.

[0113] Specific implementation of the present invention:

[0114] Referring to Figure 3 , the operation process of the method can be seen. First, the trajectory data of the vehicle is read from the road test millimeter-wave radar, and the area where it is located and the trajectory set are divided and determined; then the data of the area adjacent to the target vehicle trajectory is obtained to ensure overall consideration during matching and stitching; then different matching and stitching methods are selected for different areas, a multi-point matching method is used for the overlapping area, and the trajectory set is updated according to the following rules.

[0115] |point k,time,front -point j,time,back |≤Δtime

[0116] distance k,j ≤dist

[0117] |point k,speed,front -point j,speed,back |≤Δspeed

[0118] |point k,length,front -point j,length,back |≤Δlength

[0119] For the trajectory stitching of the non-overlapping area, an improved linear trend model is used for filling.

[0120] n=(trace begin,back -trace end,front ) / interval

[0121]

[0122] point t+n =pa t +npb t

[0123] After that, in order to avoid the deviation between the filled points and the previous historical trajectory, Kalman filtering is used for processing. On this basis, the extended Hungarian algorithm is used for matching and splicing. The transformation rules of the extended Hungarian algorithm are as follows:

[0124] maxDist = MAX(dist i,j ) + aMIN(dist i,j )

[0125] maxDiff = MAX(|lenDiff i,j |) + bMIN(lenDiff i,j |)

[0126]

[0127] Refer to Figure 4 , it can be seen the dynamic determination of the set to which the trajectory belongs. As shown in the figure, the current time is t; the allowable delay time of the millimeter-wave radar trajectory data is set to a; that is to say, at this t moment, all the data before the t - a moment have been confirmed to be received; the time determination threshold for the disappearance of the trajectory is represented as b; then the determination condition for the trajectory to enter the sudden disappearance set SDC is trace v,end + b < t - a, where trace v,end represents the last arrival time of the trajectory trace of vehicle v. No trajectory data of vehicle v is received for a continuous duration of b after the trace v,end moment. Similarly, if the allowable delay time is set to a', that is, at the current t moment, the trajectory data before the t - a' moment have all been received. In this case, the tracking window in the figure can represent the time interval for trajectory splicing. In a dynamic tracking window, further analysis yields the sudden appearance set SAC of the trajectory. Then the determination condition for the trajectory to enter the sudden appearance set SAC of the trajectory is:

[0128]

[0129] This formula defines three specific conditions, that is, first, it is judged that the trajectory trace v of the current vehicle v does not belong to the overlapping area set LC; secondly, the relationship between the time interval length of the current vehicle v in the dynamic tracking window and the time interval for trajectory splicing is judged; finally, the relationship between the start time of the trajectory of the current vehicle v in the dynamic tracking window and the sudden appearance threshold of the trajectory is judged.

[0130] Refer to Figure 5, for trajectory filling in the case of non-overlapping area trajectory splicing. The left and right sides in the figure respectively represent the detection areas of two different millimeter-wave radars. Each area has multiple dots, representing the point information of the detected vehicle trajectories. Among them, the black dots represent the detected vehicles, and the hollow dots represent the vehicles that may exist but have not been detected by the millimeter-wave radar within the range. This figure intuitively shows the movement of vehicles between the millimeter-wave radar groups. And when the vehicle enters and leaves the detection area, it may suddenly disappear or suddenly appear, which means that the appearance and disappearance of the trajectory may be separated by a relatively long distance. Therefore, the above-mentioned linear trend prediction model is used for filling in this process.

[0131] Refer to Figure 6 , in order to process arbitrarily disappearing trajectories, a method based on predictive filtering is adopted for completion. When trajectory data is missing, first, using historical points and motion laws, a filtering algorithm is used to predict the disappearing trajectory to infer its possible motion state and trajectory. This process involves dynamic modeling of the trajectory and evaluating the accuracy of the prediction by calculating the relative distance between the predicted position and the actual position. Secondly, the obtained relative distance is used as a weight and incorporated into the subsequent matching process. The introduction of the weight is mainly to improve the accuracy of matching, so that when performing trajectory matching, those trajectory pairs with relatively small relative distances can be given priority. Therefore, the extended Hungarian algorithm mentioned above is applied. After extension, this algorithm can further optimize the result based on considering the weight. In this way, not only can the problem of trajectory missing be effectively processed, but also the overall accuracy and reliability can be improved in the matching. Finally, these steps together constitute a complete trajectory prediction and matching framework, which can achieve efficient trajectory analysis and processing in a complex dynamic environment.

[0132] Refer to Figure 7 , this is the box plot of the vehicle trajectory splicing test results of the present invention on an actual road equipped with three millimeter-wave radars. Through this box plot, the effect of the present invention on the trajectory splicing length distribution can be intuitively shown. In this box plot, the box represents the middle 50% data of the trajectory length, and the line indicates the median. It is easy to see that before splicing, the trajectory length distributions within the ranges of the three radars are relatively scattered and the lengths are generally short. After being processed by the splicing method, the trajectory lengths increase significantly and the distributions are more concentrated. This fully shows that the method in this chapter can effectively connect the fragmented trajectories in different radar ranges, making the trajectory information more complete and continuous, and at the same time verifying the effectiveness of this method in improving the accuracy of target tracking and recognition.

[0133] A wide-area trajectory stitching method based on a millimeter-wave radar group provided by the present invention, its detailed steps and technical innovations are directed to applications in fields such as intelligent transportation systems and autonomous driving assistance programs. The embodiments are as follows:

[0134] 1. Set up the scenario: Install multiple millimeter-wave radar devices on a road with heavy traffic. Install one millimeter-wave radar device every 250 m on the roadside, and a total of three devices are installed. The three devices operate synchronously to capture real-time vehicle data.

[0135] 2. Data processing and trajectory set division: Filter and denoise the original vehicle trajectory information received by the radar group, and distinguish the trajectory information obtained by multiple millimeter-wave radars according to the trajectory area division logic of the present invention to clarify their respective combinations, laying a data foundation for subsequent stitching and matching in each area.

[0136] 3. Multi-point matching in the overlapping area: Perform multi-point matching on the trajectory information in the overlapping area through the matching rules (formulas) given by the present invention, and perform corresponding operations on the trajectory information that matches successfully or fails to match, and update the corresponding trajectory sets.

[0137] 4. Trajectory stitching in the non-overlapping area: This step is mainly for stitching the trajectories in the non-overlapping area, but also includes the trajectory data that fails to match in the overlapping area. On the one hand, an improved linear trend model is used to fill in the trajectory gaps, and on the other hand, combined with the uncertainty between the disappearance and appearance of the trajectories, an extended Hungarian matching algorithm is used for stitching.

[0138] In order to verify the effectiveness and applicability of the wide-area trajectory stitching method based on the millimeter-wave radar group described in the present invention, an experiment was conducted by setting up three millimeter-wave radars every 250 meters on a certain road in Qingdao, Shandong Province. The dataset used was the vehicle trajectory data obtained by millimeter-wave radar tracking. Three millimeter-wave radars were deployed at three locations, and the acquisition duration of the three radars was about 20 minutes. A total of 173 effective trajectories were obtained by the first radar, 165 effective trajectories were obtained by the second radar, and 169 effective trajectories were obtained by the third radar. The specific detection situation and trajectory information are shown in the following table.

[0139]

[0140] The method described in the present invention was selected for comparison with the Kalman filter radar global tracking algorithm (KF) and the multiple hypothesis tracking algorithm (MHT). The experiment selected two indicators, the mean absolute error (MAE) and the root mean square error (RMSE), to measure the error between the millimeter-wave radar vehicle trajectory stitched by this method and the actual vehicle trajectory. Among them, MAE can reflect the actual error situation between the estimated stitched trajectory and the actual trajectory, while RMSE can measure the deviation degree between the estimated stitched trajectory and the actual trajectory.

[0141] The trajectory stitching algorithm of the present invention is adaptively set. The tracking window duration is 4 s, a = 2 s, b = 2 s, and the vehicle trajectory is tracked almost in real time. 20 minutes of data is collected, including large vehicles, small vehicles, and non-motor vehicles. The number of small vehicles is the largest. Both scenarios of dense vehicles and sparse vehicles are collected. The comparison test results of the method of the present invention with KF and MHT are shown in the following table.

[0142]

[0143] From the experimental results, compared with the traditional Kalman filtering method and multi-hypothesis tracking algorithm, the method of the present invention performs better in both MAE and RMSE metrics. A smaller MAE value indicates the overall error of the method in this paper, while a smaller RMSE value means it is more effective in the case of larger errors or the presence of outliers, with smaller overall fluctuations. Overall, it shows that the method in the paper has better performance in the accuracy of vehicle behavior tracking or target positioning, can effectively reduce the positioning error and error fluctuations, and improve the robustness of the system.

[0144] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated designed hardware. Those of ordinary skill in the art can understand that the above devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code is provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and their modules of the present invention can be implemented by hardware circuits of programmable hardware devices such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips and transistors, or programmable logic devices such as field programmable gate arrays, and can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software such as firmware.

[0145] The above is only the specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention by those skilled in the art within the disclosed technical scope of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A wide-area vehicle trajectory stitching method based on a millimeter-wave radar group, characterized in that, Including the following steps: 1) Arrange multiple millimeter-wave radars on the road and obtain the trajectory information of the vehicle. Each millimeter-wave radar detects the targets within its front fan-shaped area and obtains the target's id, relative lateral offset, relative longitudinal offset, relative speed, relative acceleration, and timestamp information. 2) Classify the vehicle trajectories within a wide area, including identifying vehicles that suddenly disappear, suddenly appear, enter the detection area, enter the non-overlapping area, enter the overlapping area, and leave the detection area. 3) Divide the millimeter-wave radar detection area into the initial entry into the radar detection area, non-overlapping area, front overlapping area, rear overlapping area, and leaving the radar detection area, and divide the obtained trajectories into the sudden disappearance set SDC, sudden appearance set SAC, overlapping area set LC, and global tracking set GTC. 4) Perform multi-point matching in the overlapping area between adjacent millimeter-wave radars, and update the corresponding vehicle trajectory information in the corresponding set. If the matching is successful in the overlapping area, remove the corresponding vehicle record from the sudden disappearance set SDC; if the matching is successful in both the front overlapping set and the rear overlapping set, modify the corresponding trajectory in the global tracking set GTC; for vehicles that do not match successfully, add them to the sudden appearance set SAC according to the preset rules. 5) Stitch the target trajectories in the non-overlapping area, including filling in the missing trajectory data, and using the Kalman filter to process the deviation between the predicted point and the historical trajectory, and construct an extended Hungarian matching algorithm for stitching based on the relative distance and the vehicle length transformation value.

2. The method according to claim 1, wherein The rules for judging vehicle trajectory matching in step 4) are set based on multiple conditions, including the relative position, speed difference, acceleration difference, and timestamp difference of the target vehicle.

3. The wide-area vehicle trajectory stitching method based on a millimeter-wave radar group as described in claim 1, wherein The basis for determining the area to which the vehicle trajectory belongs is as follows: For the first radar: For the last radar: For the remaining radars: In the above formula, inLap represents the area where the millimeter-wave radar is located; the values 1, 2, 3, 4, and 5 represent the rear overlapping area, non-overlapping area, front overlapping area, leaving the radar detection area, and initial entry into the radar detection area respectively; x represents the coordinates of the target vehicle in the detection area; L1 and L2 represent the start and end points of the detection area respectively; S1, S2, and S3 represent the detection area distance threshold, overlapping area distance threshold, and leaving the detection area distance threshold respectively.

4. The wide-area vehicle trajectory stitching method based on a millimeter-wave radar group as claimed in claim 1, wherein The LC in the overlap region i,front , LC i,back The matching rules are: |point k,time,front -point j,time,back |≤Δtime distance k,j ≤ dist |point k,speed,front -point j,speed,back |≤Δspeed |point k,length,front -point j,length,back |≤Δlength Judgment is made by setting conditions of time, distance, speed and length, where point k,time,front represents the timestamp corresponding to the k-th trajectory on a certain trajectory in the front overlapping area; time represents the timestamp corresponding to the j-th trajectory on a certain trajectory in the rear overlapping area; point j,time,back represents the threshold for determining that two timestamps are close enough; point k,speed,front represents the speed corresponding to the k-th trajectory on a certain trajectory in the front overlapping area; point j,speed,back represents the speed corresponding to the j-th trajectory on a certain trajectory in the rear overlapping area; speed represents the threshold for determining that two speed values are close enough; point k,length,front represents the vehicle body length collected for the k-th trajectory on a certain trajectory in the front overlapping area; point j,length,back represents the vehicle body length collected for the j-th trajectory on a certain trajectory in the rear overlapping area; Δlength represents the threshold for determining that two vehicle body lengths are close enough; distance k,j represents the distance between the k-th trajectory point and the j-th trajectory point; dist represents the threshold for determining that the distance between two vehicles is close enough.

5. The wide-area vehicle trajectory stitching method based on a millimeter-wave radar group as claimed in claim 1, wherein The improved linear trend model is: n = (trace begin,back - trace end,front ) / interval point t+n = pa t + npb t where n is the number of data to be filled between the disappearance and appearance of the trajectory; trace begin,back and trace end,back represent the start time of the appearance trajectory and the end time of the disappearance trajectory respectively; interval represents the time interval between data; is the moving average of the current time t; point i is the trajectory point at the i-th moment; is the second moving average; pa t and pb t are the smoothing coefficients.

6. The wide-area vehicle trajectory stitching method based on a millimeter-wave radar group as claimed in claim 1, wherein The relative distance between the trajectories is calculated through the last point of the trajectory obtained based on filling and Kalman filtering, that is, pointNew t+n The calculation is carried out, and the distance is in the Mars coordinate system.

7. The wide-area vehicle trajectory stitching method based on a millimeter-wave radar group as claimed in claim 1, wherein The transformation rules of the extended Hungarian matching algorithm are expressed as: maxDist = MAXdist i,j + aMIN(dist i,j ) maxDiff = MAX|lenDiff i,j |+bMIN(lenDiff i,j |) where maxDist represents the maximum distance between the disappearance set and the appearance set plus an offset; maxDiff represents the maximum vehicle length difference between the disappearance set and the appearance set plus an offset; weight i,j represents the matching weight between the disappearance set and the appearance set; where the parameters a and b are both offset parameters, c is the weight parameter of the distance, and e is the weight parameter of the vehicle length difference.

8. A wide-area vehicle trajectory stitching system based on a millimeter-wave radar group for implementing the wide-area vehicle trajectory stitching method based on a millimeter-wave radar group as described in any one of claims 1-7, characterized in that, The wide-area vehicle trajectory stitching system based on a millimeter-wave radar group includes: A trajectory information acquisition module for obtaining the vehicle's trajectory information through a preset millimeter-wave radar group on the road; A classification module for classifying vehicle trajectories within a wide area; A determination module for determining the area and set to which the vehicle trajectory belongs; A matching module for performing multi-point matching in the overlapping area; A stitching module for stitching the target trajectories in the non-overlapping area.

9. A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor is caused to execute the steps of the method for wide-area vehicle trajectory stitching based on a millimeter-wave radar group as described in any one of claims 1-7.

10. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the wide-area vehicle trajectory stitching system based on a millimeter-wave radar group as described in claim 8.

Images