A forward millimeter wave radar low speed cross target detection method and device

Abstract

The application discloses a forward millimeter wave radar low-speed cross-target detection method and equipment. The method comprises the following steps: after a millimeter wave radar system completes current frame echo collection and two-dimensional fast Fourier transform, reading a distance-Doppler diagram of the current frame; calculating a main clutter velocity channel corresponding to a stationary target; dividing the distance-Doppler diagram into a plurality of sub-diagrams along a distance dimension; setting a search space according to the main clutter velocity channel; sequentially traversing each sub-diagram; searching and obtaining a target sampling point and a weight thereof; traversing the target sampling point in descending order of the weight; screening out a candidate target point according to a preset screening condition; obtaining position information of an optimal center point and M points around the optimal center point; extracting distance channel and velocity channel information of the optimal center point and the M points around the optimal center point according to the position information; and performing cross-target detection. The application avoids dynamic and static separation misjudgment, and can output a real lateral velocity of a target.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, specifically to a method and device for detecting low-speed cross-traversing targets using forward millimeter-wave radar. Background Technology

[0002] For target detection using vehicle-mounted millimeter-wave radar, the industry's commonly adopted approach is a point cloud generation and motion / static separation process based on Constant False Alarm Rate (CFAR) detection, followed by target tracking. This process includes: performing a two-dimensional Fast Fourier Transform (FFT) on the radar echo signal to obtain a Range-Doppler (RD) map. The horizontal axis of this map represents the range channel, and the vertical axis represents the velocity channel, with each coordinate point representing the echo intensity. The CFAR algorithm is then applied to the RD map, performing threshold discrimination on a cell-by-cell basis to detect cells with intensities higher than the noise floor, generating the original detection points. Velocity calculations are performed on all point clouds to obtain their radial velocities. Subsequently, based on the geometric relationship between the radial velocity and the vehicle's speed, the point clouds are classified into "stationary point clouds" or "moving point clouds." The classified points are then clustered, aggregating adjacent point clouds into clusters to form a target point cloud list. The point cloud labeled as "moving point cloud" is then tracked to output the target's motion information (distance, velocity, etc.); the point cloud labeled as "stationary point cloud" is either filtered out or used for static target construction.

[0003] In existing technical solutions, the Target Cloud Detection (CFAR) stage can detect point clouds containing tangentially moving targets from the RD map. However, in the "motion-stationary separation" step after point cloud generation, when a target is moving tangentially, its direction of motion is almost perpendicular to the radar normal, resulting in a very small radial velocity component in the tangential direction. Therefore, point clouds that originally belong to moving targets have radial velocity values ​​similar to those of stationary targets (such as guardrails, cones, etc.), and are incorrectly identified as "stationary point clouds" during motion-stationary separation, thus being filtered out or forming stationary targets, and unable to enter the moving target tracking and velocity estimation stages.

[0004] Based on the above analysis, the existing technical solutions have the following inherent defects caused by the physical principles of millimeter-wave radar: When a target is moving tangentially, its point cloud can be detected normally by CFAR. However, because the radial velocity has a very small lateral component, these point clouds are mistakenly identified as stationary target point clouds (such as guardrails, cones, etc.) and filtered out during motion-to-stationary separation. Ultimately, the radar cannot identify these point clouds as moving targets, and therefore cannot output the correct velocity information.

[0005] For forward-facing radar, the most common tangential motion scenarios include low-speed targets crossing the road, "ghost pedestrians" emerging from roadside parking spaces, and targets crossing with a small azimuth angle. In these highly dangerous scenarios, existing technologies may misidentify the target's motion state, causing autonomous driving systems to fail to predict collision risks in a timely manner. Therefore, there is an urgent need for a method that can ensure these tangentially moving targets, which are mistakenly identified as stationary, are correctly identified as moving targets and output the correct speed information. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and device for detecting low-speed cross-traversing targets using forward millimeter-wave radar.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a method for detecting low-speed traversing targets using forward-facing millimeter-wave radar, comprising: Step 1: After the millimeter-wave radar system completes the echo acquisition and two-dimensional fast Fourier transform of the current frame, it reads the range-Doppler image of the current frame. Step 2: Obtain the vehicle's speed and, in conjunction with the operating parameters of the millimeter-wave radar, calculate the main clutter velocity channel corresponding to the stationary target; Step 3: Divide the distance-Doppler map into multiple sub-maps along the distance dimension; Step 4: Set the search space according to the main clutter velocity channel, traverse each sub-graph in turn, and search to obtain the target sampling point and its weight; Step 5: Traverse the target sampling points in descending order of weight, and select candidate target points according to preset filtering conditions. Take the first selected candidate target point as the candidate result of the current subgraph, and take the first selected candidate target point as the center point. Record the position information of the M target sampling points around the center point. The subsequent selected candidate target points are determined according to the preset selection rules to determine whether to replace the existing candidate results, so as to obtain the optimal center point and the position information of the M points around it. Step 6: Extract the distance and velocity channel information of the optimal center point and its surrounding M points based on the location information, and realize the cross-target detection based on the extracted distance and velocity channel information.

[0008] Furthermore, the search space is set as follows: Based on the main clutter velocity channel, the upward offset is no more than 5 velocity channels, and the downward extension extends to a search space width of 20 velocity channels.

[0009] Furthermore, the target sampling points are obtained based on particle filter search, specifically as follows: Equally weighted particles are randomly scattered within the search space. Iterative state prediction, weight updates based on distance-Doppler echo intensity, and resampling are performed to converge the particles to the peak position of the echo energy amplitude within the search space. The converged particles are then used as the target sampling points.

[0010] Furthermore, the specific selection rules are as follows: Compare the distances between the subsequently selected candidate target points and the existing candidate results with the candidate results of the previous frame. If the distance between the subsequently selected candidate target point and the candidate results of the previous frame is less than the distance between the existing candidate results and the candidate results of the previous frame, or if the distance between the subsequently selected candidate target point and the candidate results of the previous frame is equal to the distance between the existing candidate results and the candidate results of the previous frame and the echo intensity of the subsequently selected candidate target point is higher than that of the existing candidate results, then the subsequently selected candidate target point replaces the existing candidate results; otherwise, the existing candidate results are not replaced.

[0011] Furthermore, when performing the target sampling point search, each subgraph is traversed sequentially from far to near.

[0012] Furthermore, the filtering criteria are as follows: The distance-Doppler echo intensity corresponding to the coordinates of the target sampling point is greater than a preset echo intensity threshold, and the number of target sampling points with intensity greater than the echo intensity threshold in an n×n neighborhood centered on the target sampling point exceeds N.

[0013] Furthermore, the echo intensity threshold is 70.

[0014] Furthermore, the value of n is 3, the value of M is 8, and the value of N is 4.

[0015] In a second aspect, the present invention provides a forward millimeter-wave radar low-speed cross-target detection device, comprising a storage medium and a processor, wherein the storage medium stores a computer program, which, when executed by the processor, is used to implement the above-described method.

[0016] Beneficial effects: 1. This invention directly extracts the point cloud that crosses the target from the RD map and outputs it in the form of 9 point clusters. At the same time, it is specially marked so that the subsequent processing can identify it as a moving target, avoid misjudgment of motion-static separation, enter the moving target tracking module, and finally output the target's true lateral velocity. 2. The search space of this invention is limited to an asymmetric region with no more than 5 channels above the main clutter velocity channel and 20 channels below. This setting is consistent with the characteristic of low-speed moving targets traversing across the Doppler offset downwards, while also restricting the entry of clutter from above. Attached Figure Description

[0017] Figure 1This is a schematic diagram of a forward millimeter-wave radar low-speed cross-target detection method according to an embodiment of the present invention. Detailed Implementation

[0018] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0019] like Figure 1 As shown, this embodiment of the invention provides a method for detecting low-speed cross-traversing targets using forward-facing millimeter-wave radar, comprising: Step 1: After the millimeter-wave radar system completes the current frame echo acquisition and two-dimensional Fast Fourier Transform (FFT), it reads the range-Doppler (RD) map. Specifically, the millimeter-wave radar system first performs a range-dimensional FFT to calculate the target range, then performs a Doppler-dimensional FFT to calculate the target relative velocity. The two FFTs are combined to form a two-dimensional FFT, and the results constitute the RD map. In the RD map: X-axis: distance channel, each column corresponds to a distance range, representing how far the target is from the radar; Y-axis: velocity channel, each row corresponds to a Doppler velocity, representing the radial velocity of the target relative to the radar; Z-axis: Echo intensity, corresponding to the signal amplitude at each coordinate point, usually expressed as brightness / color. The stronger the object's reflection, the higher the value.

[0020] Step 2: Obtain the vehicle's speed and, combined with the millimeter-wave radar's operating parameters (speed resolution), calculate the main clutter speed channel. Specifically, since the direction of motion of a target traversing the radar is perpendicular to the radar normal, its radial velocity component in the tangential direction is relatively small. The target's speed channel is usually adjacent to the main clutter speed channel corresponding to a stationary target. The target speed obtained by the millimeter-wave radar is the relative speed with the vehicle as the reference point; that is, the relative speed of a stationary target is the negative value of the vehicle's speed. Therefore, the main clutter speed channel can be obtained by dividing the vehicle's speed by the speed resolution. The vehicle's speed can be obtained via the CAN bus.

[0021] Step 3: Divide the distance-Doppler map into multiple sub-maps along the distance dimension (X-axis). Specifically, the preferred number of sub-maps is 4, and they are equally divided. However, this is not limited to this; 2, 3, 8, etc., are also possible. The segmentation method can also be non-uniform, for example, using finer-grained segmentation in common target-crossing distance intervals.

[0022] Step 4: Based on the main clutter velocity channel, set the search space, and sequentially traverse each sub-image to obtain the target sampling points and their weights. Specifically, the search space is set as follows: Using the main clutter velocity channel as a reference, the search space is shifted upwards by no more than 5 velocity channels and downwards by 20 velocity channels. For example, 3 channels are taken upwards and 16 channels are taken downwards from the main clutter velocity channel, totaling 20 channels. This asymmetric region design, while ensuring detection capability, maximizes the suppression of overhead clutter, thus reducing false alarms.

[0023] When performing target sampling point search, each subgraph is traversed sequentially from farthest to nearest. Longer-distance subgraphs are processed first to detect distant targets crossing the map earlier, allowing the autonomous driving system more braking reaction time.

[0024] The target sampling points mentioned above can be obtained based on particle filter search, as follows: Equally weighted particles are randomly scattered within the search space. Iterative processes include state prediction, weight updates based on distance-Doppler echo intensity, and resampling, until the particles converge to the peak position of the echo energy amplitude within the search space. The converged particles are then used as the target sampling points. However, this method is not limited to this; using genetic algorithms or simulated annealing algorithms can also achieve global energy extremum search within the search space to find the target sampling points.

[0025] Step 5: Traverse the target sampling points in descending order of weight, and select candidate target points according to the preset filtering conditions. Take the first selected candidate target point as the candidate result of the current subgraph, and take the first selected candidate target point as the center point. Record the position information of the M target sampling points around the center point. The subsequent selected candidate target points are determined according to the preset selection rules to determine whether to replace the existing candidate results, so as to obtain the optimal center point and the position information of the M points around it.

[0026] Specifically, the above screening criteria are as follows: The distance-Doppler echo intensity corresponding to the coordinates of the target sampling point is greater than a preset echo intensity threshold, and the number of target sampling points with intensity greater than the echo intensity threshold in an n×n neighborhood centered on the target sampling point exceeds N. The echo intensity threshold is preferably 70, but it can also be an adaptive relative threshold based on the subgraph noise basis. n is 3, and N is 4. However, it is not limited to these values; n can also be 5, 7, etc., and N can also be 3, 5, or a percentage condition.

[0027] The specific selection rules are as follows: Compare the distances between the subsequently selected candidate target points and the existing candidate results with the candidate results of the previous frame. If the distance between the subsequently selected candidate target point and the candidate results of the previous frame is less than the distance between the existing candidate results and the candidate results of the previous frame, or if the distance between the subsequently selected candidate target point and the candidate results of the previous frame is equal to the distance between the existing candidate results and the candidate results of the previous frame and the echo intensity of the subsequently selected candidate target point is higher than that of the existing candidate results, then the subsequently selected candidate target point replaces the existing candidate results; otherwise, the existing candidate results are not replaced.

[0028] Step 6: Extract the distance and velocity channel information of the optimal center point and its surrounding M points based on the location information, and perform cross-traversing target detection based on the extracted distance and velocity channel information. The preferred value for M is 8. The extracted distance and velocity channel information of the 9 points is sent to the CFAR detection module and then output as a "low-speed cross-traversing target special point cloud". These point clouds have special tags in the target list and are sent to the subsequent moving target tracking module along with the regular CFAR detection point clouds. After receiving these point clouds, the tracking module can perform targeted trajectory tracking based on the tags and output the target's motion information (distance, velocity, etc.). If no target is detected in the current frame, this type of special point cloud is not output. Furthermore, during the next frame processing, in the particle filter initialization stage, more particles are preferentially seeded near the distance channel where the previous frame's detection result is located to ensure continuous extraction of the same target.

[0029] Based on the above embodiments, those skilled in the art can easily understand that the present invention also provides a forward millimeter-wave radar low-speed cross-target detection device, including a storage medium and a processor, wherein the storage medium stores a computer program, and the computer program is executed by the processor to implement the above-described method.

[0030] The above description is merely a preferred embodiment of the present invention. It should be noted that for those skilled in the art, other parts not specifically described are existing technology or common knowledge. Several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting low-speed traversing targets using forward-facing millimeter-wave radar, characterized in that, include: Step 1: After the millimeter-wave radar system completes the echo acquisition and two-dimensional fast Fourier transform of the current frame, it reads the range-Doppler image of the current frame. Step 2: Obtain the vehicle's speed and, in conjunction with the operating parameters of the millimeter-wave radar, calculate the main clutter velocity channel corresponding to the stationary target; Step 3: Divide the distance-Doppler map into multiple sub-maps along the distance dimension; Step 4: Set the search space according to the main clutter velocity channel, traverse each sub-graph in turn, and search to obtain the target sampling point and its weight; Step 5: Traverse the target sampling points in descending order of weight, and select candidate target points according to preset filtering conditions. Take the first selected candidate target point as the candidate result of the current subgraph, and take the first selected candidate target point as the center point. Record the position information of the M target sampling points around the center point. The subsequent selected candidate target points are determined according to the preset selection rules to determine whether to replace the existing candidate results, so as to obtain the optimal center point and the position information of the M points around it. Step 6: Extract the distance and velocity channel information of the optimal center point and its surrounding M points based on the location information, and realize the cross-target detection based on the extracted distance and velocity channel information.

2. The method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 1, characterized in that, The search space is set as follows: Based on the main clutter velocity channel, the upward offset is no more than 5 velocity channels, and the downward extension extends to a search space width of 20 velocity channels.

3. The method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 1, characterized in that, The target sampling points are obtained based on particle filter search, as follows: Equally weighted particles are randomly scattered within the search space. Iterative state prediction, weight updates based on distance-Doppler echo intensity, and resampling are performed to converge the particles to the peak position of the echo energy amplitude within the search space. The converged particles are then used as the target sampling points.

4. The method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 1, characterized in that, The specific selection rules are as follows: Compare the distances between the subsequently selected candidate target points and the existing candidate results with the candidate results of the previous frame. If the distance between the subsequently selected candidate target points and the candidate results of the previous frame is less than the distance between the existing candidate results and the candidate results of the previous frame, or if the distance between the subsequently selected candidate target points and the candidate results of the previous frame is equal to the distance between the existing candidate results and the candidate results of the previous frame and the echo intensity of the subsequently selected candidate target points is higher than that of the existing candidate results, then replace the existing candidate results with the subsequently selected candidate target points. Otherwise, do not replace existing candidate results.

5. The method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 1, characterized in that, When performing the target sampling point search, each subgraph is traversed sequentially from farthest to nearest.

6. The method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 1, characterized in that, The filtering criteria are as follows: The distance-Doppler echo intensity corresponding to the coordinates of the target sampling point is greater than a preset echo intensity threshold, and the number of target sampling points with intensity greater than the echo intensity threshold in an n×n neighborhood centered on the target sampling point exceeds N.

7. A method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 6, characterized in that, The echo intensity threshold is 70.

8. A method for detecting low-speed traversing targets using forward-facing millimeter-wave radar according to claim 6, characterized in that, The value of n is 3, the value of M is 8, and the value of N is 4.

9. A forward-facing millimeter-wave radar low-speed traversing target detection device, comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that, When the computer program is executed by a processor, it is used to implement the method described in any one of claims 1-8.

Images