Millimeter-wave radar vehicle type classification method

Abstract

The present invention discloses a millimeter-wave radar vehicle type classification method, which includes: using a millimeter-wave radar to detect a target to be detected to obtain the position information and velocity information of each measured trace, and performing preprocessing; constructing an elliptical gate model, and clustering each trace from bottom to top through the elliptical gate model, the position information and velocity information of each trace to obtain a preliminary classification result; tracking multiple target traces in the preliminary classification result to form a target track; determining an associated target combination according to the elliptical gate model and the velocity information of the tracking target corresponding to the target track, and calculating the potential large vehicle probability of each associated target in the associated target combination through a multi-frame sliding window, merging the associated targets that meet the preset conditions, and determining them as the large vehicle type. The present invention realizes vehicle type classification through vehicle type clustering and target tracking results, thereby effectively improving the accuracy of vehicle type classification.

Description

Technical Field

[0001] The present invention relates to the technical field of radar target detection, and particularly to a millimeter-wave radar vehicle type classification method. Background Art

[0002] Millimeter-wave radar has the characteristics of real-time ranging and speed measurement for multiple targets and all-weather operation, and has high detection accuracy, strong anti-interference performance, small size and low installation cost. At present, it is widely used in the transportation field. In the actual road traffic scenario, vehicle type is important road traffic information, and vehicle type classification based on millimeter-wave radar is still a difficult point. Especially for the detection of large vehicles, the radar may detect multiple target points.

[0003] The existing vehicle type classification methods mainly cluster the target vehicles. However, the existing clustering methods mainly classify the targets by judging the length and width or the number of scattering points of the clustering clusters. Such methods have a high dependence on the clustering results. When the point distribution is uneven, it is very easy to miscluster in some areas, thus affecting the accuracy of vehicle type classification. Summary of the Invention

[0004] The present invention aims to provide a millimeter-wave radar vehicle type classification method to improve the accuracy of vehicle type classification. To achieve the above object, the present invention is realized through the following technical solutions:

[0005] A millimeter-wave radar vehicle type classification method includes:

[0006] Step S1: Detect a target to be detected by using a millimeter-wave radar to obtain the position information and speed information of each trace corresponding to the target to be detected at the current moment, and preprocess the obtained traces;

[0007] Step S2: Determine the prior information of the elliptical gate, construct an elliptical gate model through the prior information of the elliptical gate, and cluster each of the traces from bottom to top through the elliptical gate model, the position information and the speed information of each trace to obtain a preliminary classification result;

[0008] Step S3: Track multiple target traces in the preliminary classification result to form a target track;

[0009] Step S4: Determine an associated target combination according to the elliptical gate model and the speed information of the tracking target corresponding to the target track, calculate the potential large vehicle probability of each associated target in the associated target combination through a multi-frame sliding window, merge each associated target that meets the preset conditions, and determine the type of the merged tracking target as the large vehicle type.

[0010] Optionally, in the step S1, the step of preprocessing the obtained traces includes:

[0011] Eliminate the traces and velocity-detected-zero traces outside the effective detection range of the millimeter-wave radar.

[0012] Optionally, the prior information of the elliptical gate in step S2 includes the elliptical minor-axis parameter and the elliptical major-axis parameter, and the elliptical gate model is represented by the following formula:

[0013]

[0014] In the formula, E x is the elliptical minor-axis parameter, E y is the elliptical major-axis parameter, (x0, y0) is the coordinate of the center point of the ellipse, and (x, y) is the trace coordinate within the elliptical gate.

[0015] Optionally, the elliptical gate model includes a car elliptical gate model and a large vehicle elliptical gate model. In step S2, the step of clustering each of the traces from bottom to top through the elliptical gate model, the position information, and the velocity information of each trace includes:

[0016] Step S21: Determine whether the trace falls within the elliptical gate range corresponding to the car elliptical gate model according to the position information of the trace;

[0017] Step S22: Traverse and calculate, and merge all the traces that fall within the elliptical gate range corresponding to the car elliptical gate model and whose velocity difference is less than or equal to the first preset velocity threshold to obtain the preliminary classification result.

[0018] Optionally, step S3 includes:

[0019] Step S31: When the current frame is the starting moment, use the clustered target traces as the starting points of the target tracks;

[0020] Step S32: When the current frame is not the starting moment, perform nearest neighbor association between the multi-target traces of the current frame and the predicted traces of the target tracks in the previous frame, and update according to the association result;

[0021] Among them, when the target trace is successfully associated with the predicted trace of the target track in the previous frame, perform Kalman filtering using the target trace and the predicted trace to obtain the current frame update value of the corresponding target track.

[0022] Optionally, in step S4, the step of determining the associated target combination according to the elliptical gate model and the velocity information of the tracking target corresponding to the target track includes:

[0023] Step S41: When the current frame is at the starting moment, determine whether the tracking target falls within the elliptical gate range corresponding to the large vehicle elliptical gate model according to the position information of the tracking target; combine all the tracking targets that fall within the elliptical gate range corresponding to the large vehicle elliptical gate model and whose speed difference is less than or equal to the second preset speed threshold to obtain the associated target combination, and calculate the center point position information and speed information of the associated target combination;

[0024] Step S42: When the current frame is not at the starting moment, use the predicted point of the center point of the associated target combination in the previous frame as the center of the elliptical gate, update all the tracking targets that fall within the elliptical gate range corresponding to the center of the elliptical gate and whose speed difference from the predicted point is less than or equal to the second preset speed threshold to the associated target combination, and calculate the center point position information and speed information of the updated associated target combination.

[0025] Optionally, within the multi-frame sliding window, when the number of valid frames of the associated target in the corresponding associated target combination reaches the preset number of frames, it is determined that the preset condition is satisfied; where, when the associated target is in the corresponding associated target combination, calculate the potential large vehicle probability of the associated target, and when the potential large vehicle probability is greater than the preset probability value, it is recorded that the preset condition is satisfied.

[0026] Optionally, the potential large vehicle probability is expressed by the following formula:

[0027]

[0028] In the formula, P1 is P(large vehicle|successful association of target within combination), which represents the probability that the associated target is successfully associated with the associated target combination and is a large vehicle, P2 is P(successful association of target within combination), which represents the probability that the associated target is successfully associated with the associated target combination, P3 is the probability that the associated target is associated with the associated target combination unsuccessfully and is a large vehicle, P4 is the probability that the associated target is associated with the associated target combination unsuccessfully, where P1 and P3 are prior probabilities.

[0029] The present invention has at least the following technical effects:

[0030] (1) The present invention adopts a double - layer clustering framework. First, based on the bottom - up hierarchical clustering idea, the traces measured by the radar are clustered to obtain a preliminary classification result. On the basis of target tracking, a multi - frame sliding window method is used to calculate the probability that a target is a large vehicle. Targets that meet the preset conditions are merged to obtain the final classification result, realizing the classification of cars and large vehicles. Compared with the prior art that only judges the vehicle type based on the clustering result, the present invention comprehensively considers the results of clustering and tracking to achieve vehicle type classification, and can avoid the situation where a large vehicle is split into multiple targets due to the failure of trace clustering, which helps to improve the quality of target tracking and thus can improve the accuracy of vehicle type classification;

[0031] (2) The present invention associates the information at the current moment with the information at the historical moment, obtains the probability that a target is a large vehicle, and determines the final classification result, thereby overcoming the defect of the prior art that relies on single - frame information to identify the vehicle type and making the classification result more reasonable and reliable.

[0032] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. Brief Description of the Drawings

[0033] Figure 1 It is a flowchart of the millimeter - wave radar vehicle type classification method provided by an embodiment of the present invention;

[0034] Figure 2A It is a schematic diagram of the measurement traces for millimeter - wave radar vehicle type classification provided by an embodiment of the present invention;

[0035] Figure 2B It is a schematic diagram of the preliminary clustering result of millimeter - wave radar vehicle type classification provided by an embodiment of the present invention;

[0036] Figure 3 It is a flowchart of target tracking for the millimeter - wave radar vehicle type classification method provided by an embodiment of the present invention;

[0037] Figure 4 It is a schematic diagram of a real - world traffic scene provided by an embodiment of the present invention;

[0038] Figures 5 - 7 It is a schematic diagram of vehicle type classification results under different numbers of frames provided by an embodiment of the present invention. Detailed Description of the Embodiment

[0039] The following details this embodiment. The examples of the embodiment are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0040] The millimeter-wave radar vehicle type classification method of this embodiment will be described below with reference to the accompanying drawings.

[0041] Figure 1 The flowchart of the millimeter-wave radar vehicle type classification method provided by an embodiment of the present invention is as follows. Figure 1 As shown, the method includes:

[0042] Step S1: Detect the target to be detected using a millimeter-wave radar to obtain the position information and velocity information of each trace corresponding to the target to be detected at the current moment, and preprocess the obtained traces.

[0043] Among them, the step of preprocessing the obtained traces includes: removing the traces outside the effective detection range of the millimeter-wave radar and the traces with zero velocity detection.

[0044] Specifically, trace preprocessing includes position constraint and filtering of stationary traces, that is, removing the traces outside the effective detection range of the radar and the traces with zero velocity.

[0045] In this embodiment, the millimeter-wave radar can be used to detect the target to be detected, obtain information such as the position and velocity of the traces within a period of time, convert them into spatial coordinates and velocity information, and remove the traces outside the effective detection range of the radar (y > 500m, |x| > 20m) and the traces with a velocity of 0m / s, that is, only consider the moving targets on the road, and organize the trace information into a structure format. The content of the target structure includes the lateral distance, longitudinal distance, lateral velocity, and longitudinal velocity of each frame of trace.

[0046] Step S2: Determine the prior information of the elliptical gate, construct an elliptical gate model through the prior information of the elliptical gate, and cluster each trace from bottom to top through the elliptical gate model, the position information, and the velocity information of each trace to obtain a preliminary classification result.

[0047] Among them, the prior information of the elliptical gate includes the elliptical minor axis parameter and the elliptical major axis parameter. The elliptical gate model is expressed by the following formula:

[0048]

[0049] In the formula, E x is the elliptical minor axis parameter, E y is the elliptical major axis parameter, (x0, y0) is the coordinate of the center point of the ellipse, and (x, y) is the coordinate of the trace inside the elliptical gate.

[0050] Among them, when the elliptical minor axis parameter and the elliptical major axis parameter are the parameters corresponding to the car elliptical gate, the obtained is the car elliptical gate model. When the elliptical minor axis parameter and the elliptical major axis parameter are the parameters corresponding to the large vehicle elliptical gate, the obtained is the large vehicle elliptical gate model.

[0051] In this embodiment, the step of clustering each trace from bottom to top through the elliptical gate model, the position information and velocity information of each trace includes:

[0052] Step S21: Determine whether the trace falls within the elliptical gate range corresponding to the trolley elliptical gate model according to the position information of the trace.

[0053] Step S22: Through traversal calculation, merge all the traces that fall within the elliptical gate range corresponding to the trolley elliptical gate model and whose velocity difference is less than or equal to the first preset velocity threshold to obtain a preliminary classification result.

[0054] Specifically, the prior information of the elliptical gate can be used to cluster the traces using the idea of bottom-up hierarchical clustering. For example, set the prior information of the trolley elliptical gate (E x =1.6, E y =5.5) to construct the trolley elliptical gate model. Then, as Figures 2A - 2B shown, take all the traces at the current moment as different classes. By traversing all the traces, merge the classes that fall within the trolley elliptical gate range corresponding to this trolley elliptical gate model and satisfy the velocity difference Δv≤6m / s, and take the average value of the coordinates and velocities as the target information after clustering, and then assign the target trolley type.

[0055] Step S3: Track multiple target traces in the preliminary classification result to form a target track.

[0056] Among them, step S3 includes:

[0057] When the current frame is the starting moment, take the clustered target trace as the starting point of the target track.

[0058] When the current frame is not the starting moment, perform nearest neighbor association between the multi-target traces in the current frame and the predicted traces of the target track in the previous frame, and update according to the association result.

[0059] Among them, when the target trace is successfully associated with the predicted trace of the target track in the previous frame, perform Kalman filtering using the target trace and the predicted trace to obtain the updated value of the current frame of the corresponding target track.

[0060] As Figure 3As shown, if the current frame is the starting moment, the clustered traces can be used as the starting points of the tracks; if the current frame is not the starting moment, the multi-target traces in the current frame can be associated with the predicted trace points of the tracks obtained in the previous frame through the nearest neighbor method, and the tracks can be updated according to the association results. If a certain track has no associated trace points for multiple consecutive frames, the track is deleted. If the trace points in the current frame are not associated with the tracks, the trace points are used as the starting points of new tracks. If the tracks are successfully associated with the trace points in the current frame, the Kalman filter is used with the trace points in the current frame and the predicted points of the previous frame of the tracks to obtain the updated value of the current frame of the track, and the target state of the next frame is predicted.

[0061] In this embodiment, the tracking of multiple target traces is realized according to the clustering result to form the content of the track, which specifically includes track start, track update, and track termination. In this embodiment, the logical method can be used to start the track for the data in the clustered target structure, and the track information is organized into a track structure, which includes track number, position, speed, the count of frames without track association, and Kalman filter parameters. In this embodiment, track update is to predict the radar measurement points of the current frame for all tracks in the track structure of the previous frame using the Kalman filter method of the uniform linear motion model, calculate the association between the predicted value and the trace points in the target structure of the current frame. When the trace points pass through the distance gate and speed gate set for a certain track, record the distance between them, and use the nearest neighbor association method to associate the trace points with the nearest track and perform Kalman filter update.

[0062] Furthermore, traverse all tracks in the track structure. If each track successfully associates with trace points in the current frame, use the above nearest neighbor association method and Kalman filter to update the track; if a track has no associated trace points in the current frame, use the predicted value to update the track; for the trace points of the remaining unassociated tracks in the current frame, use them as the starting points of new tracks. Among them, track termination means that a certain track has no associated trace points within multiple consecutive frames. At this time, it is considered that the target no longer exists, and the track is deleted.

[0063] Step S4: Determine the associated target combination according to the elliptical gate model and the speed information of the tracking target corresponding to the target track, calculate the potential large vehicle probability of each associated target in the associated target combination through a multi-frame sliding window, merge the associated targets that meet the preset conditions, and determine the type of the merged tracking target as the large vehicle type.

[0064] Among them, the step of determining the associated target combination according to the elliptical gate model and the speed information of the tracking target corresponding to the target track includes:

[0065] Step S41: When the current frame is at the starting moment, determine whether the tracking target falls within the elliptical gate range corresponding to the large vehicle elliptical gate model according to the position information of the tracking target; combine all the tracking targets that fall within the elliptical gate range corresponding to the large vehicle elliptical gate model and whose speed difference is less than or equal to the second preset speed threshold to obtain an associated target combination, and calculate the center point position information and speed information of the associated target combination.

[0066] Step S42: When the current frame is not at the starting moment, use the predicted point of the center point of the associated target combination in the previous frame as the center of the elliptical gate, update all the tracking targets that fall within the elliptical gate range corresponding to the center of the elliptical gate and whose speed difference from the predicted point is less than or equal to the second preset speed threshold to the associated target combination, and calculate the center point position information and speed information of the updated associated target combination.

[0067] In this embodiment, the associated target combination is a set of targets that satisfy the large vehicle elliptical gate model and the speed difference within each frame based on the position and speed of the tracking target. If the current frame is at the starting moment, assign the same combination label to the targets that meet the association conditions, and calculate the center point position and speed of the associated target combination; if the current frame is not at the starting moment, use the predicted points of the center points of each associated target combination in the previous frame as the center of the elliptical gate, and determine the targets in the current frame that are within the elliptical gate range corresponding to this center and satisfy the speed difference, then update the track numbers within the associated target combination, recalculate the center point information of the associated target combination, and predict the state of the next frame.

[0068] In this embodiment, within the multi-frame sliding window, when the number of valid frames of the associated target in the corresponding associated target combination reaches the preset number of frames, it can be determined that the preset conditions are met; among them, the specific method for determining that the preset conditions are met is that when the associated target is in the corresponding associated target combination, calculate the potential large vehicle probability of the associated target, and when the potential large vehicle probability is greater than the preset probability value, it is recorded that the preset conditions are met.

[0069] Among them, the potential large vehicle probability is calculated using Bayes' formula:

[0070]

[0071] In the formula, P1 is P(target successfully associated within the large vehicle combination), which represents the probability that the associated target is successfully associated with the associated target combination and is a large vehicle, P2 is P(target successfully associated within the combination), which represents the probability that the associated target is successfully associated with the associated target combination, P3 is the probability that the associated target is associated with the associated target combination unsuccessfully and is a large vehicle, and P4 is the probability that the associated target is associated with the associated target combination unsuccessfully. Among them, P1 and P3 are prior probabilities, which are obtained through a large amount of data statistics and analysis.

[0072] Specifically, the sliding window can be set to N frames. If the number of frames in which the associated target exists in the associated target combination reaches M frames, that is, the potential large vehicle probability of the associated target is greater than the preset probability value of 85%, the associated targets that meet these conditions are merged, and the type of the tracked target after merging is determined as the large vehicle type.

[0073] For example, the interval between each frame of the millimeter-wave radar is about 70 milliseconds. To obtain the vehicle type classification result in a short time, the sliding window N can be set to 29 frames, that is, the sliding window is about 2s. Then, through formula calculation, it is obtained that when the associated target in the associated target combination is successfully associated 8 frames within the 29-frame sliding window and the potential large vehicle probability exceeds 85%, the target type can be determined as a large vehicle.

[0074] As a specific example, Figure 4 is a schematic diagram of a real traffic scene. In this scene, as Figures 5 - 7 shown, there is 1 large truck as a moving target and 1 passenger car as a stationary target. It should be noted that the present invention does not consider stationary targets (stationary dots have been excluded), and only tracks moving targets. At the 339th frame, the large truck starts to enter the radar detection range. Due to the uneven distribution of scatter points, it is mis-clustered into 3 small cars. According to the Bayesian formula calculation, the potential large vehicle probabilities of the targets with track numbers 145 and 146 are 35.56%; when it is the 353rd frame, the large truck has completely entered the radar detection range, the potential large vehicle probabilities of the targets with track numbers 146 and 150 increase to 82.96%, and the potential large vehicle probability of the target with track number 145 reaches 85.33%; when it is the 355th frame, the potential large vehicle probabilities of all 3 targets exceed 85%. At this time, the 3 targets can be merged into one target and determined as the large vehicle type.

[0075] In summary, the present invention adopts a double-layer clustering framework, clusters the radar measurement dots based on the bottom-up hierarchical clustering idea and the prior information of the elliptical wave gate to obtain the preliminary target classification result. Then, based on the target clustering and tracking results, it judges the targets that meet the association within each frame according to the elliptical wave gate and speed information, and assigns the same combination label. It calculates the potential large vehicle probability of the targets within each combination in a multi-frame sliding window manner, merges the targets whose potential large vehicle probability within each combination exceeds the threshold, and determines the large vehicle as the final classification result of the target, so that the present invention can effectively improve the accuracy of vehicle type classification.

[0076] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0077] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A method for classifying vehicle models by millimeter-wave radar, characterized in that, Including: Step S1: Use a millimeter-wave radar to detect a target to be detected, so as to obtain the position information and velocity information of each trace corresponding to the target to be detected at the current moment, and preprocess the obtained traces. Step S2: Determine the prior information of the elliptical gate, construct an elliptical gate model through the prior information of the elliptical gate, and cluster each of the traces from bottom to top through the elliptical gate model, the position information, and the velocity information of each trace to obtain a preliminary classification result; the elliptical gate model includes a car elliptical gate model and a large vehicle elliptical gate model. The step of clustering each of the traces from bottom to top through the elliptical gate model, the position information, and the velocity information of each trace includes: Step S21: Determine whether the trace falls within the elliptical gate range corresponding to the car elliptical gate model according to the position information of the trace. Step S22: Through traversal calculation, merge all the traces that fall within the elliptical gate range corresponding to the car elliptical gate model and whose velocity difference is less than or equal to the first preset velocity threshold to obtain the preliminary classification result. Step S3: Track multiple target traces in the preliminary classification result to form a target track, specifically including: Step S31: When the current frame is the starting moment, use the clustered target traces as the starting points of the target track. Step S32: When the current frame is not the starting moment, perform nearest neighbor association between the multi-target traces in the current frame and the predicted traces of the target track in the previous frame, and update according to the association result. Among them, when the target trace is successfully associated with the predicted trace of the target track in the previous frame, perform Kalman filtering using the target trace and the predicted trace to obtain the current frame update value of the corresponding target track. Step S4: Determine an associated target combination according to the elliptical gate model and the velocity information of the tracking target corresponding to the target track, calculate the potential large vehicle probability of each associated target in the associated target combination through a multi-frame sliding window, merge each associated target that meets the preset conditions, and determine the type of the merged tracking target as the large vehicle type. The step of determining an associated target combination according to the elliptical gate model and the velocity information of the tracking target corresponding to the target track includes: Step S41: When the current frame is the starting moment, determine whether the tracking target falls within the elliptical gate range corresponding to the large vehicle elliptical gate model according to the position information of the tracking target; combine all the tracking targets that fall within the elliptical gate range corresponding to the large vehicle elliptical gate model and whose velocity difference is less than or equal to the second preset velocity threshold to obtain the associated target combination, and calculate the center point position information and velocity information of the associated target combination. Step S42: When the current frame is not the starting moment, use the predicted point of the center point of the associated target combination in the previous frame as the center of the elliptical gate. Update all tracking targets that fall within the elliptical gate range corresponding to the center of the elliptical gate and whose velocity difference from the predicted point is less than or equal to the second preset velocity threshold to the associated target combination, and calculate the position information and velocity information of the center point of the updated associated target combination; In the multi-frame sliding window, when the number of valid frames of the associated target in the corresponding associated target combination reaches the preset number of frames, it is determined that the preset condition is satisfied; among them, when the associated target is in the corresponding associated target combination, calculate the potential large vehicle probability of the associated target, and when the potential large vehicle probability is greater than the preset probability value, it is recorded that the preset condition is satisfied.

2. The millimeter-wave radar vehicle type classification method according to claim 1, wherein In the step S1, the steps of preprocessing the acquired traces include: Eliminate the traces outside the effective detection range of the millimeter-wave radar and the traces with zero velocity detection.

3. The millimeter-wave radar vehicle type classification method according to claim 2, wherein, The prior information of the elliptical gate in the step S2 includes the elliptical minor axis parameter and the elliptical major axis parameter, and the elliptical gate model is expressed by the following formula: where E x is the minor axis parameter of the ellipse, E y is the major axis parameter of the ellipse, (x0, y0) are the coordinates of the center point of the ellipse, and (x, y) are the coordinates of the point trace within the elliptical gate.

4. The millimeter-wave radar vehicle type classification method according to claim 3, characterized in that The potential large vehicle probability is expressed by the following formula: In the formula, P1 represents the probability that the associated target is successfully associated with the associated target combination and is a large vehicle, P2 represents the probability that the associated target is successfully associated with the associated target combination, P3 is the probability that the associated target is unsuccessfully associated with the associated target combination and is a large vehicle, P4 is the probability that the associated target is unsuccessfully associated with the associated target combination, where P1 and P3 are prior probabilities.

Images