A method for splicing highway vehicle trajectories based on radar data
By preprocessing radar data and optimizing trajectory stitching algorithms, the problem of discontinuous vehicle trajectory stitching on highways by millimeter-wave radar has been solved, enabling accurate identification and continuous monitoring of vehicle trajectories, and making it suitable for scenarios with multiple ramps and curves.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHONGLU JIAOKE TECHNOLOGY CO LTD
- Filing Date
- 2023-01-31
- Publication Date
- 2026-08-04
AI Technical Summary
In existing technologies, millimeter-wave radar lacks consideration for lane changes when tracking vehicle trajectories on highways, resulting in multiple radars being unable to accurately stitch together the trajectory of the same vehicle, leading to discontinuous vehicle trajectory data.
By acquiring vehicle trajectory data from multiple radars in real time, preprocessing, dividing by lane, and unifying the coordinate system, the system can determine the overlapping parts of the trajectories or lane change situations. The system then uses a motion state model and an LSTM model to predict the position of trajectory points, thus achieving trajectory stitching.
It achieves accurate identification and continuous stitching of vehicle trajectories, is suitable for complex scenarios, improves the accuracy and real-time performance of trajectory stitching, and supports vehicle operation status identification and illegal behavior monitoring.
Smart Images
Figure CN116008990B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of vehicle trajectory stitching technology, and in particular to a method for stitching highway vehicle trajectories based on radar data. Background Technology
[0002] Intelligent Transportation Systems (ITS) are crucial technologies for alleviating traffic problems and ensuring traffic safety, and traffic information collection is the foundation and guarantee for the effective functioning of ITS. With the continuous advancement of traffic detection technology, various methods for collecting traffic information are increasingly emerging. Currently, the widely used vehicle trajectory recognition method both domestically and internationally is camera-based. However, cameras have low accuracy in recognizing small, distant targets, leading to missing vehicle trajectories. Furthermore, cameras are not sensitive to vehicle speed, and vehicle trajectories acquired by cameras on different road sections cannot be effectively stitched together based on vehicle motion characteristics. These two aspects result in discontinuous vehicle trajectories between different devices when using cameras as sensors. Nevertheless, acquiring long-range continuous vehicle trajectories is of great value for vehicle operation status identification, continuous monitoring of vehicle violations, driving behavior analysis, and the development of intelligent traffic management. Therefore, there is an urgent need to propose a vehicle trajectory stitching method based on roadside sensors.
[0003] As the hardware cost of millimeter-wave radar gradually decreases, some high-precision millimeter-wave radars used in military applications are being open-sourced for civilian use. Currently, they account for a significant portion of automotive collision avoidance sensors; according to IHS data, millimeter-wave / microwave radar combined with cameras accounts for 70% of automotive collision avoidance sensors. Millimeter-wave radar is characterized by its short wavelength, wide bandwidth (large frequency range), and strong penetration capabilities. These characteristics constitute its advantages, giving it enormous potential for application in the transportation sector. The collection and analysis of vehicle trajectory information in road areas can be based on millimeter-wave radar data.
[0004] However, lane changes are common on highways. When millimeter-wave radar tracks a vehicle's trajectory, it does not take into account lane changes. There is a problem that multiple millimeter-wave radars cannot accurately stitch together the trajectory of the same vehicle, resulting in discontinuous vehicle trajectory data when multiple millimeter-wave radars detect the same vehicle target. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to provide a method for stitching together highway vehicle trajectories based on radar data, which effectively solves the problems in the background art.
[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for stitching highway vehicle trajectories based on radar data, comprising:
[0007] Real-time acquisition of vehicle trajectory data from multiple radars;
[0008] Preprocessing abnormal data points in vehicle trajectory data;
[0009] The preprocessed vehicle trajectory point data is divided by lane;
[0010] The coordinate system is used to determine the positions of road vehicles within the field of view of each radar.
[0011] Iterate through the vehicle trajectory data processed by two adjacent radars to determine whether there is any overlap between the corresponding time periods of the two trajectories in the same lane.
[0012] If there is an overlap, determine the unified trajectory point of the two trajectories according to the matching conditions and perform trajectory splicing; otherwise, further determine whether a lane change has occurred.
[0013] If a lane change occurs, the motion state model is used to predict the position of the next timestamp of the trajectory point. When the predicted trajectory point and the actual trajectory point are in the same lane and the distance between them is 15m, the two points are determined to be the same trajectory point, and the two trajectories to which the two points belong are spliced together.
[0014] Furthermore, the vehicle trajectory data is obtained through radar latitude and longitude, illumination angle, target distance, vehicle latitude and longitude coordinates, direction angle, trajectory and speed.
[0015] Furthermore, the preprocessing involves removing noise from the radar data that is not related to vehicle trajectories. The specific steps are as follows:
[0016] Based on the density clustering and velocity of the trajectory points, it is determined whether the trajectory points themselves are abnormal;
[0017] Based on whether the position and velocity of adjacent timestamp trajectory points are within a reasonable range, determine and analyze whether adjacent trajectory points are abnormal.
[0018] If there is abnormal data in the trajectory point itself or in adjacent trajectory points, the abnormal trajectory point data will be deleted.
[0019] Furthermore, the specific steps for dividing trajectory points by lane are as follows:
[0020] Oversampling occurred when k-means clustering was used on the preprocessed radar data.
[0021] Based on the data obtained after oversampling, the trajectory points are laterally clustered into n lanes;
[0022] Obtain the centerline of each lane of the road, and thus obtain the coordinate range corresponding to each lane;
[0023] Based on the coordinate range corresponding to each lane, each trajectory point is divided by lane.
[0024] Furthermore, segmented clustering is used when clustering lanes laterally, and a transitional sampling method based on clustering is used to oversample lanes with few vehicle trajectory points so that the trajectory data of each lane is the same.
[0025] Furthermore, the steps to unify the coordinates of the trajectory points acquired by each radar are as follows:
[0026] Given the northward angles of two adjacent upstream and downstream radars, rotate the downstream radar coordinate system to be consistent with the upstream radar coordinate system, and calculate the relative positions of the origins of the upstream and downstream radar coordinate systems based on the relative positions of the two adjacent upstream and downstream radars.
[0027] Based on the relative position, the rotated downstream radar coordinate system is transformed into the upstream radar coordinate system, and the trajectory points within the downstream radar's field of view are transformed into coordinates of the upstream radar coordinate system.
[0028] Furthermore, the matching condition is as follows: if the lanes are the same and the timestamps differ by 1 second, and the distance between the trajectory points differs by 5m, then the two points are considered to be the same trajectory points.
[0029] Furthermore, when there is no overlap between the corresponding time periods of two trajectories on the same lane, if the speed of the trajectory points is greater than 20 km / h, it is considered that the vehicle has changed lanes.
[0030] Furthermore, when a lane change occurs, if the trajectory data is dense, a velocity model is used to predict the position of the trajectory point at the next timestamp, specifically:
[0031] y pre =y pos +y speed ×Δt
[0032] Among them, y pos The endpoint of the first half of the trajectory is y. speed Let y be the velocity at the end of the first half of the trajectory, Δt be the time difference between the end of the first half of the trajectory and the beginning of the second half of the trajectory, and y be the velocity at the end of the first half of the trajectory. pre This represents the predicted trajectory point location.
[0033] Furthermore, when a lane change occurs, if the trajectory data is sparse, a single-layer LSTM model is used to predict the position of the trajectory point at the next timestamp based on the position at historical time.
[0034] The beneficial effects of this invention are as follows: it optimizes the traditional highway vehicle trajectory stitching algorithm in many aspects, uses radar data to identify vehicle trajectories more accurately, uses a clustering-based oversampling method to eliminate radar data noise in the radar data preprocessing part, and adds judgment on highway lanes such as ramps in the trajectory stitching algorithm part to make the matching and stitching more accurate. The overall algorithm can achieve real-time and high efficiency.
[0035] This invention preprocesses, clusters, and predicts millimeter-wave radar data based on clustering and machine learning algorithms to identify vehicle trajectory data. It then matches vehicle trajectories identified by multiple radars and stitches together the trajectories of individual vehicles into long-range continuous vehicle trajectory data. This enables subsequent functions such as vehicle operating status identification, continuous monitoring of illegal behavior, and driving behavior analysis. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a flowchart of a highway vehicle trajectory stitching method based on radar data in an embodiment of the present invention;
[0038] Figure 2 This is a schematic diagram of data preprocessing in an embodiment of the present invention;
[0039] Figure 3 This is a schematic diagram of the lane division structure of radar unit 1 data in an embodiment of the present invention;
[0040] Figure 4 This is a schematic diagram of the lane division structure of radar unit 2 data in an embodiment of the present invention;
[0041] Figure 5 This is a schematic diagram of the lane division structure of radar unit 3 data in an embodiment of the present invention. Detailed Implementation
[0042] 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.
[0043] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.
[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0045] In this invention, as Figure 1 The method for stitching highway vehicle trajectories based on radar data, as shown, includes:
[0046] S1: Real-time acquisition of vehicle trajectory data from multiple radars; When the millimeter-wave radar collects vehicle trajectory information in the road area, its vehicle trajectory data is obtained through radar position (latitude and longitude), radar illumination angle, and target distance. The vehicle trajectory data includes latitude and longitude coordinates, direction angle, vehicle trajectory, and vehicle speed.
[0047] S2: Preprocessing of outlier data points in vehicle trajectory data. Preprocessing involves cleaning the trajectory data points to remove outliers, such as targets like wind and trees. Figure 2 As shown, star-shaped markers represent anomalous data points. When deleting noise from non-vehicle trajectories in radar data, based on the latitude and longitude of trajectory points and / or the average speed corresponding to the trajectory points detected by millimeter-wave radar, erroneous trajectories within the trajectory data are identified. This requires analyzing the trajectory point itself and its adjacent trajectory points. The specific steps are as follows:
[0048] Based on the density clustering and velocity of the trajectory points, it is determined whether the trajectory points themselves are abnormal; that is, when a trajectory point is far away from the dense area of trajectory points and / or the longitudinal velocity of the trajectory point is greater than 180km / h or less than 40km / h, it is considered to be an abnormal trajectory point.
[0049] Based on whether the position and speed of adjacent timestamp trajectory points are within a reasonable range, we can determine whether adjacent trajectory points are abnormal. When the position of a trajectory point exceeds the position range of the preceding and following trajectory points, and the speed of the trajectory point differs from the speed of the preceding and following trajectory points by 20 km / h, then the trajectory point is also considered an abnormal data point.
[0050] If a trajectory point itself or an adjacent trajectory point is abnormal, the abnormal trajectory point data will be deleted.
[0051] S3: Divide the preprocessed vehicle trajectory point data by lane;
[0052] Cluster analysis is performed on the cleaned data to analyze the number of lanes and the lanes in which vehicles are located. Lateral clustering is performed in the direction parallel to the cross-section of the road surface. Lateral clustering is the initial stable point clustering. The purpose of lateral stable point clustering is to first determine the lane data based on the vehicle trajectory, and use this as a reference point for subsequent vehicle trajectory acquisition, thereby determining the lane in which the vehicle is located.
[0053] S4: Perform coordinate transformation on the positions of road vehicles within the field of view of a single radar, and unify the coordinates of adjacent radars into a unified coordinate system;
[0054] S5: Traverse the vehicle trajectory data processed by two adjacent radars and determine whether there is any overlap between the corresponding time periods of the two trajectories in the same lane.
[0055] S6: If there is an overlap, determine the unified trajectory point of the two trajectories according to the matching conditions and perform trajectory splicing; otherwise, further determine whether a lane change has occurred.
[0056] When two trajectories on the same lane overlap within the same time period, according to the matching conditions, if the lanes are the same, the timestamps differ by 1 second, and the distance between the trajectory points differs by 5 meters, then these two points are considered to be the same trajectory point. However, when two trajectories on the same lane do not overlap within the same time period, if the speed of the trajectory point is greater than 20 km / h, then the vehicle is considered to have changed lanes.
[0057] S7: If a lane change occurs, the motion state model is used to predict the position of the next timestamp of the trajectory point. When the predicted trajectory point and the actual trajectory point are in the same lane and the distance between them is 15m, the two points are determined to be the same trajectory point, and the two trajectories to which the two points belong are spliced together.
[0058] Compared to similar vehicle trajectory stitching algorithms based on millimeter-wave radar, the trajectory stitching method proposed in this invention is applicable to various highways and complex scenarios with multiple ramps and curves. It optimizes traditional highway vehicle trajectory stitching algorithms in several aspects, using radar data for more accurate vehicle trajectory identification. In the radar data preprocessing section, a clustering-based oversampling method is used to eliminate radar data noise. In the trajectory stitching algorithm, the judgment of highway lanes such as ramps is added, making the matching and stitching more accurate.
[0059] The radar establishes a Cartesian coordinate system, while roads have curvature; therefore, segmented clustering is required during lateral clustering. In weaving zones, ramps will be treated as lanes of the main road for clustering. For example, on the Nanjing-Shanghai Expressway... Figure 3-5 As shown, the main road has 4 lanes, and there is 1 ramp in the weaving area, so the total number of lanes is determined to be 5. The specific steps for dividing the trajectory points according to the lanes are as follows:
[0060] Oversampling occurred when k-means clustering was used on the preprocessed radar data.
[0061] Based on the data obtained after oversampling, the trajectory points were laterally clustered into 5 lanes after the trajectory was divided into 50-meter segments.
[0062] Obtain the centerline of each lane of the road, and further obtain the coordinate range corresponding to each lane;
[0063] Based on the coordinate range corresponding to each lane, each trajectory point is divided by lane.
[0064] The lateral clustering of this invention obtains the center point of each lane in a road segment, obtains the lane width based on the coordinate range of the left lane of each lane, and determines the lane where each trajectory point is located by combining the lateral coordinates of each trajectory point.
[0065] Because the number of vehicles in some lanes is much smaller than in others (such as truck lanes), for lanes with few vehicle trajectories, to avoid them being overlooked during clustering due to insufficient trajectories, an oversampling method based on clustering is used to oversample these lanes. This ensures that the trajectory data for each lane is identical, thereby guaranteeing the accuracy and stability of the subsequently acquired lane alignments and improving the computational stability of the method in this invention. Oversampling achieves sample balance by increasing (copying) the number of minority class samples in the classification. For example, if there are 100 samples in class A and 10 in class B, then to ensure the balance between classes A and B, class B samples can be copied to make them identical to class A.
[0066] This invention requires matching and stitching vehicle trajectory data read from multiple radars. However, the coordinate systems of the data from different radars are not consistent, making matching difficult. Therefore, it is necessary to unify the coordinate systems or vehicle trajectory data representation formats of each radar to a certain extent. The steps to unify the coordinates of the trajectory points acquired by each radar are as follows:
[0067] Given the northward angles of two adjacent upstream and downstream radars, rotate the downstream radar coordinate system to be consistent with the upstream radar coordinate system, and calculate the relative positions of the origins of the upstream and downstream radar coordinate systems based on the relative positions of the two adjacent upstream and downstream radars.
[0068] Based on the relative position, the rotated downstream radar coordinate system is transformed into the upstream radar coordinate system, and the trajectory points within the downstream radar's field of view are transformed into coordinates of the upstream radar coordinate system.
[0069] When a vehicle changes lanes, if the trajectory data is dense, a speed model is used to predict the position of the trajectory point at the next timestamp, specifically: y pre =y pos +y speed ×Δt
[0070] Among them, y pos The endpoint of the first half of the trajectory is y. speed Let y be the velocity at the end of the first half of the trajectory, Δt be the time difference between the end of the first half of the trajectory and the beginning of the second half of the trajectory, and y be the velocity at the end of the first half of the trajectory. pre This represents the predicted trajectory point location.
[0071] If the trajectory data is sparse, a single-layer LSTM model is used to predict the location of the trajectory point at the next timestamp based on its historical location. Specifically:
[0072] Step 1. Set up the training set: 1200 vehicle trajectories containing more than 4 trajectory points.
[0073] Step 2. Set up the test set: 300 vehicle trajectories
[0074] Step 3. After training, the model's accuracy in predicting location stabilizes at 80%.
[0075] Compared to traditional vehicle trajectory stitching algorithms that use cameras as sensors, this invention offers higher accuracy and stronger real-time performance. Based on clustering and machine learning algorithms, it preprocesses, clusters, and predicts millimeter-wave radar data to identify vehicle trajectory data. It then matches the vehicle trajectories identified by multiple radars, stitching together the trajectories of individual vehicles into long-range continuous vehicle trajectory data. This enables subsequent functions such as vehicle operating status identification, continuous monitoring of illegal behavior, and driving behavior analysis.
[0076] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.
Claims
1. A method for stitching together vehicle trajectories on highways based on radar data, characterized in that, include: Real-time acquisition of vehicle trajectory data from multiple radars; Preprocessing abnormal data points in vehicle trajectory data; The preprocessed vehicle trajectory point data is divided by lane; The coordinate system is used to determine the positions of road vehicles within the field of view of each radar. Iterate through the vehicle trajectory data processed by two adjacent radars to determine whether there is any overlap between the corresponding time periods of the two trajectories in the same lane. If there is an overlap, determine the unified trajectory point of the two trajectories according to the matching conditions and perform trajectory splicing; otherwise, further determine whether a lane change has occurred. If a lane change occurs, the motion state model is used to predict the position of the next timestamp of the trajectory point. When the predicted trajectory point and the actual trajectory point are in the same lane and the distance is 15m apart, the two points are determined to be the same trajectory point, and the two trajectories to which the two points belong are spliced together. When a lane change occurs, if the trajectory data is dense, the velocity model is used to predict the position of the trajectory point at the next timestamp, specifically: and pre / and pos +and speed ×Δt Among them, y pos The endpoint of the first half of the trajectory is y. speed Let y be the velocity at the end of the first half of the trajectory, Δt be the time difference between the end of the first half of the trajectory and the beginning of the second half of the trajectory, and y be the velocity at the end of the first half of the trajectory. pre The predicted trajectory point positions; When a lane change occurs, if the trajectory data is sparse, a single-layer LSTM model is used to predict the location of the trajectory point at the next timestamp based on the location at historical time.
2. The method for stitching highway vehicle trajectories based on radar data according to claim 1, characterized in that, The vehicle trajectory data is obtained through radar latitude and longitude, illumination angle, target distance, vehicle latitude and longitude coordinates, direction angle, trajectory and speed.
3. The method for stitching highway vehicle trajectories based on radar data according to claim 1, characterized in that, The preprocessing involves removing noise from the radar data that is not related to vehicle trajectories. The specific steps are as follows: Based on the density clustering and velocity of the trajectory points, it is determined whether the trajectory points themselves are abnormal; Based on whether the position and velocity of adjacent timestamp trajectory points are within a reasonable range, determine and analyze whether adjacent trajectory points are abnormal. If there is abnormal data in the trajectory point itself or in adjacent trajectory points, the abnormal trajectory point data will be deleted.
4. The method for stitching highway vehicle trajectories based on radar data according to claim 1, characterized in that, The specific steps for dividing trajectory points by lane are as follows: Oversampling occurred when k-means clustering was used on the preprocessed radar data. Based on the data obtained after oversampling, the trajectory points are laterally clustered into n lanes; Obtain the centerline of each lane of the road, and thus obtain the coordinate range corresponding to each lane; Based on the coordinate range corresponding to each lane, each trajectory point is divided by lane.
5. The method for stitching highway vehicle trajectories based on radar data according to claim 4, characterized in that, When clustering lanes laterally, segmented clustering is used, and an oversampling method based on clustering is used to oversample lanes with few vehicle trajectory points so that the trajectory data of each lane is the same.
6. The method for stitching highway vehicle trajectories based on radar data according to claim 1, characterized in that, The steps to unify the coordinates of the trajectory points acquired by each radar are as follows: Given the northward angles of two adjacent upstream and downstream radars, rotate the downstream radar coordinate system to be consistent with the upstream radar coordinate system, and calculate the relative positions of the origins of the upstream and downstream radar coordinate systems based on the relative positions of the two adjacent upstream and downstream radars. Based on the relative position, the rotated downstream radar coordinate system is transformed into the upstream radar coordinate system, and the trajectory points within the downstream radar's field of view are transformed into coordinates of the upstream radar coordinate system.
7. The method for stitching highway vehicle trajectories based on radar data according to claim 1, characterized in that, The matching conditions are as follows: if the lanes are the same and the timestamps differ by 1 second, and the distance between the trajectory points differs by 5 meters, then the two points are considered to be the same trajectory points.
8. The method for stitching highway vehicle trajectories based on radar data according to claim 1, characterized in that, When there is no overlap between the corresponding time periods of two trajectories on the same lane, if the speed of the trajectory points is greater than 20 km / h, it is considered that the vehicle has changed lanes.