A traffic multi-target association method based on spatial grouping and feature matching

Abstract

The application relates to the technical field of intelligent traffic perception, in particular to a traffic multi-target correlation method based on spatial grouping and feature matching. First, a spatial position grouping algorithm is used to pre-classify detection targets, a dynamic grouping model is established based on the spatial density distribution and grouping relationship of the targets, and the target search space is effectively reduced; then, multi-dimensional correlation metrics are calculated, a correlation matrix is constructed, the motion consistency and feature similarity of the targets are comprehensively evaluated, the optimal matching result of all the matching successful track records in the same group is recorded, the matching result is output in a delay mode, when more than 50% of the targets in the same group are all matched successfully or the target matching times exceed 10 times, the point record with the most matching times in the track is output as the current matching result. The traffic multi-target correlation method based on spatial position grouping and feature matching can significantly improve the accuracy and real-time performance of multi-target correlation by fusing the spatial distribution characteristics, motion features and delay output of the targets.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation perception technology, specifically a traffic multi-objective association method based on spatial grouping and feature matching. Background Technology

[0002] In existing technologies, multi-target association methods mainly include the traditional nearest neighbor (NN) algorithm, which associates the current observation with the target closest to the predicted location, and the joint probabilistic data association algorithm (JPDA), which calculates the weighted sum of association probabilities between all valid observations and targets. However, these methods face the following technical bottlenecks:

[0003] False association of dense targets: In densely trafficked scenarios, the traditional nearest neighbor (NN) algorithm suffers from a significantly higher false association rate because it does not consider the spatial grouping relationships between targets. Experimental data shows that when the distance between targets is less than 10 meters, the false association rate of the NN algorithm exceeds 25%.

[0004] Insufficient adaptability to dynamic environments: Existing motion model-based association methods (such as JPDA) are sensitive to changes in target maneuverability, and the association accuracy drops by more than 40% in rapid acceleration and deceleration scenarios;

[0005] Insufficient utilization of features: Traditional methods often rely on single motion features (such as velocity vectors) and ignore the synergistic effect of spatial grouping relationships between targets (such as relative position and density distribution) and multi-dimensional features (RCS, SNR). Therefore, in view of the above situation, it is urgent to develop a traffic multi-target association method based on spatial grouping and feature matching to overcome the shortcomings in current practical applications. Summary of the Invention

[0006] The purpose of this invention is to provide a traffic multi-target association method based on spatial grouping and feature matching, which is applicable to multi-target tracking and trajectory association of millimeter-wave radar in complex traffic scenarios, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A traffic multi-objective association method based on spatial grouping and feature matching includes the following steps:

[0009] S1. Real-time acquisition of point cloud data of vehicle targets via millimeter-wave radar;

[0010] S2. Preprocess the point cloud data to extract the target feature vector set, wherein the feature vector includes position, speed, lane information and heading angle;

[0011] S3. Dynamically group targets based on their spatial density distribution and lane positions;

[0012] S4. Perform hierarchical association matching on trajectory targets and point targets within the group: primary matching screening: filter valid point subsets based on position, velocity and heading angle; precise matching calculation: generate matching combinations and calculate comprehensive difference degree, and select the optimal matching scheme;

[0013] S5. Verify and output the matching results: When the preset conditions are met, update the trajectory status and output the correlation results.

[0014] As a further aspect of the present invention: in step S3, dynamic grouping specifically includes:

[0015] Spatial clustering algorithm is used, with vertical distance threshold. The lane position tolerance ΔL is used as the grouping condition, when the target satisfies and When grouped into the same group, output the set of groups. .

[0016] As a further aspect of the present invention: in step S4, the precise matching calculation includes:

[0017] For including individual trajectory targets and Grouping of point targets to generate Given a set of matching combinations, calculate the overall dissimilarity for each combination. , and select The smallest combination is the optimal match.

[0018] As a further aspect of the present invention: the overall difference is calculated as follows:

[0019] ;

[0020] in, For speed difference, and This represents the degree of positional difference.

[0021] As a further aspect of the present invention: in step S5, the preset conditions include any of the following:

[0022] (a) More than 50% of the targets within the same group were successfully matched;

[0023] (b) The cumulative number of matches for the same target is ≥10.

[0024] As a further aspect of the present invention: in step S5, the output includes sending the association result to the traffic control center via the V2X communication module.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] This invention combines target spatial grouping relationships with feature matching. Compared with traditional global association methods, this method improves the association accuracy by 18.7% in dense traffic scenarios and has stronger robustness against target occlusion and missed detection.

[0027] This invention is applicable to intelligent traffic monitoring, vehicle-road cooperative perception and other systems, and provides an efficient and reliable solution for multi-target tracking in complex scenarios. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the traffic multi-objective association method based on spatial grouping and feature matching in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the data processing flow in an embodiment of the present invention.

[0030] Figure 3 This is a flowchart illustrating the matching process between target trajectory and point traces in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram comparing the tracking results of different methods in a high-density scene according to an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0033] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0034] Please refer to 1, which provides a traffic multi-objective association method based on spatial grouping and feature matching in this embodiment of the invention. This method constructs target spatial groups; calculates the association matrix to achieve multi-dimensional association of motion, space, and features; and develops a fast association algorithm based on grouping constraints, improving the association accuracy by 17.1% compared to the traditional nearest neighbor association algorithm. A schematic diagram of the data processing flow is shown below. Figure 2 As shown, the specific content includes:

[0035] S01. Real-time point cloud data of vehicle targets is collected by millimeter-wave radar installed on the roadside;

[0036] S02. Preprocess the point cloud data, including target clustering and feature extraction, to construct a target feature vector set. ,in , including target location ,speed Lane and driving direction (Heading angle);

[0037] S03. The steps for dynamic grouping based on spatial location distribution are as follows:

[0038] The target grouping is achieved using a spatial clustering algorithm based on an improved DBSCAN, with a neighborhood radius. ,in, The longitudinal distance threshold (typically 50 meters in a highway scenario), ΔL is the lane position tolerance (typically 1 in a highway scenario), and the target is divided into K spatial groups. };

[0039] ;

[0040] ;

[0041] ;

[0042] Wherein, ΔL is the lane position tolerance (according to the driving manual, during the lane change process, the driver can only change to the adjacent lane at a time. After completing the current lane change operation and driving stably, the next lane change can be carried out. Therefore, the lane change action must have an intermediate stable state (≥3s). The number of lane changes allowed within the driving range can be calculated based on the vehicle speed).

[0043] S04. The specific steps of the target hierarchical association matching process are as follows:

[0044] Primary matching and filtering: For m trajectory targets and n point targets within a group, filter by position, velocity, and heading angle, and output a subset of valid point targets. ;

[0045] Exact match calculation: for those containing Individual target and Grouping of trajectory targets, generating Permutations and combinations, matching combinations { }; for each combination Calculate the overall difference:

[0046] ;

[0047] Select the best match ;

[0048] S05. Matching result verification and output, details are as follows:

[0049] Output conditions: More than 50% of the targets in the same group are successfully matched; the cumulative number of target matches is ≥10.

[0050] If any condition is met, output the matching result and update the trajectory status;

[0051] The correlation results are sent to the traffic control center via the V2X communication module.

[0052] Specifically, this invention relates to a traffic multi-target association method based on spatial grouping and feature matching. It employs an improved composite distance DBSCAN algorithm for target grouping. A schematic diagram of the target point and trajectory matching process is shown below. Figure 3 As shown, this specifically includes: setting a Y-axis distance threshold. =50 meters; Set the lane position difference threshold ΔL=1; When two targets meet the following conditions and When grouped together, output the grouping results. Each group Include One trajectory target;

[0053] For each group Execution: Establish a candidate matching set: Perform coarse matching between m trajectory targets and n point targets within the group, filter by position and velocity, and output a subset of valid point targets. ;

[0054] Matching and associating targets within the same group specifically includes:

[0055] 1. Construct a matching candidate set: For those containing Individual target and Grouping of trajectory targets, generating There are several permutations and combinations, each representing a possible matching scheme. ;

[0056] 2. Calculate the matching difference: For each matching scheme Calculate its overall degree of difference Basis for calculating the degree of difference:

[0057] ;

[0058] in, For speed difference, and The specific calculation method for the positional difference is as follows:

[0059] In high-speed scenarios, calculations can be performed directly using Euclidean distance and velocity difference:

[0060] ;

[0061] ;

[0062] ;

[0063] The speed of the two targets The X-axis coordinate is Y-axis coordinate .

[0064] In complex intersection scenarios, it is necessary to use the velocity cosine similarity to determine the velocity direction, and then use the Euclidean distance and velocity difference to calculate the difference:

[0065] two target velocity vectors The formula for calculating cosine similarity is:

[0066] ;

[0067] When the value is 1, it means that the two targets are in the same direction. When the two targets are in the same direction, the difference between the two targets is further calculated. If the targets cross or are in opposite directions, it is directly determined that the two targets do not match.

[0068] 3. Determine the optimal match: Select the one that makes... Minimal matching scheme Record the successfully matched point-trajectory pairs as the optimal matching result. ;

[0069] 4. Verify and output matching results, and set the conditions for successful matching:

[0070] In real-time target tracking scenarios, when vehicles are at intersections or in areas where radar coverage overlaps with highways, the target detection data reported by the radar system does not follow a spatial order principle (i.e., it is not reported according to the relative positions of the target and the radar). Therefore, the traditional First-Come-First-Served strategy will fail in such scenarios. Thus, when setting the output conditions for matching results, a delayed output should be considered. To ensure the accuracy of target matching results within a group while also considering the real-time nature of the matching results, the first condition is set as follows:

[0071] Condition 1: More than 50% of the targets within the same group are successfully matched;

[0072] This ensures that results are output when more than half of the targets are matched successfully, avoiding increased matching difficulty due to excessively long prediction times. Additionally, to prevent result distortion caused by excessively long prediction times, the prediction distance... It should not exceed 40 meters.

[0073] ;

[0074] In dense scenes, the average frame interval of this radar is 80ms. Therefore, when condition 2 is met, the output condition is also met.

[0075] Condition 2: The cumulative number of matches for the same target is ≥ n;

[0076] Therefore, when any condition is met, the final matching result is output to update the target state information in the trajectory database.

[0077] In summary, this invention ensures matching accuracy through a three-stage processing flow of grouping, matching, and verification, and by calculating multi-dimensional differences. Figure 4 This diagram illustrates the comparison of tracking results using different methods in high-density scenarios. Experiments show that even in high-density scenarios where the target spacing is less than 10 meters, this method can still maintain an association accuracy of over 92%.

[0078] It should be noted that, in this invention, although the specification describes the embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A traffic multi-objective association method based on spatial grouping and feature matching, characterized in that, Includes the following steps: S1. Real-time acquisition of point cloud data of vehicle targets via millimeter-wave radar; S2. Preprocess the point cloud data to extract the target feature vector set, wherein the feature vector includes position, speed, lane information and heading angle; S3. Dynamically group the targets based on their spatial density distribution and lane positions to generate a spatial grouping relationship diagram; S4. Perform hierarchical association matching on trajectory targets and point targets within the group: primary matching and filtering: filter valid point subsets based on position, velocity and heading angle; Precise matching calculation: Generate matching combinations and calculate the overall difference degree to select the optimal matching scheme; S5. Verify and output the matching results: When the preset conditions are met, update the trajectory status and output the association results; In step S3, dynamic grouping specifically includes: Spatial clustering algorithm is used, with vertical distance threshold. The lane position tolerance ΔL is used as the grouping condition, when the target satisfies and When grouped into the same group, output the set of groups. ; In step S4, the exact match calculation includes: For including individual trajectory targets and Grouping of point targets to generate Given a set of matching combinations, calculate the overall dissimilarity for each combination. , and select The smallest combination is the optimal match.

2. The traffic multi-objective association method based on spatial grouping and feature matching according to claim 1, characterized in that, The overall difference is calculated as follows: ； in, For speed difference, and This represents the degree of positional difference.

3. The traffic multi-objective association method based on spatial grouping and feature matching according to claim 1, characterized in that, In step S5, the preset conditions include any of the following: (a) More than 50% of the targets within the same group were successfully matched; (b) The cumulative number of matches for the same target is ≥10.

4. The traffic multi-objective association method based on spatial grouping and feature matching according to claim 1, characterized in that, In step S5, the output includes sending the correlation results to the traffic control center via the V2X communication module.

Images