Interference Filtering Method, Device, Storage Medium and Electronic Device for Millimeter-Wave Radar Detection

Abstract

The present application discloses an interference filtering method, device, storage medium and electronic device for millimeter-wave radar detection, including: obtaining point cloud data of a detection target through a millimeter-wave radar; determining the current category feature and current target distance of the point cloud based on the point cloud data; based on the current category feature and current target distance, comparing with the relationship between the category feature and target distance of normal targets under each set category of the current scene determined in advance, determining the scores of the detection target belonging to each set category; judging whether the detection target belongs to each set category based on the scores, and if the detection target does not belong to any set category, determining that the detection target is an interference target; filtering out the point cloud data of the interference target. Applying the present application can effectively identify interference targets in millimeter-wave radar detection and improve the accuracy of millimeter-wave radar detection.

Description

Technical Field

[0001] The present application relates to millimeter-wave radar technology, and particularly to an interference filtering method, device, storage medium and electronic device for millimeter-wave radar detection. Background Art

[0002] Capturing and perceiving traffic targets such as motor vehicles and non-motor vehicles is the most core processing task in the intelligent road traffic service scenario. Currently, the road monitoring system usually combines millimeter-wave radar and camera to form a radar-vision tracking system, which integrates the rich semantics of vision and the motion target perception advantage of radar. Through multi-modal data fusion for target tracking and detection and evidence collection, good road traffic monitoring and tracking effects can be achieved.

[0003] In business practice, the above-mentioned processing method combining millimeter-wave radar and camera has remarkable effects, but many problems have also been encountered. One of the more typical problems is the target misdetection problem caused by multipath in the radar. The multipath misdetection of millimeter-wave radar is the misdetected point cloud formed after the millimeter-wave is reflected once or multiple times by obstacles such as walls and ground, and the misdetected target formed by the point cloud, which will cause great interference to target perception and affect the accuracy of millimeter-wave radar detection. For example, in a tunnel scenario, Figure 1a is a schematic diagram of the positions where the radar target and point cloud are projected onto the ground. Figure 1b is a schematic diagram of the target track and point cloud positions in the radar coordinate system. Figure 1a For the two parallel trucks in the foreground, due to multipath, Figure 1b there are multiple multipath misdetected targets marked by circles in Figure 1a and the corresponding positions in are marked by arrows. It can be seen that the multipath misdetected targets form interference in the distance.

[0004] The present application provides an interference filtering method, device, storage medium and electronic device for millimeter-wave radar detection, which can effectively identify interference targets in millimeter-wave radar detection and improve the accuracy of millimeter-wave radar detection.

[0005] To achieve the above object, the present application adopts the following technical solutions:

[0006] An interference filtering method for millimeter-wave radar detection includes:

[0007] Obtaining point cloud data of a detection target through a millimeter-wave radar;

[0008] Determining the current category feature and current target distance of the point cloud based on the point cloud data;

[0009] Based on the current category features and the current target distance, compare with the relationship between the category features and the target distance of normal targets under each set category of the current scene determined in advance, and determine the scores of the detection target belonging to each set category respectively;

[0010] Based on the scores, determine whether the detection target belongs to each set category. If the detection target does not belong to any set category, determine that the detection target is an interference target;

[0011] Filter out the point cloud data of the interference target.

[0012] Preferably, the category features include: the number of points feature of the target point cloud, the signal-to-noise ratio SNR feature of the target point cloud, the velocity consistency feature of the target point cloud, and / or the radar cross section RCS feature.

[0013] Preferably, the method for determining the relationship between the category features and the target distance of normal targets under any set category of the current scene includes:

[0014] Perform statistical analysis on the category feature values of the target point clouds of multiple normal targets under the any set category at multiple target distances, and fit to obtain the relationship curve between the category feature values of the normal target point clouds and the target distance under the any set category.

[0015] Preferably, when there are multiple category features, the scores for determining that the detection target belongs to each set category respectively include:

[0016] For any category feature, based on the relationship between the any category feature of normal targets under each set category and the target distance, determine the scores corresponding to the any category feature of the detection target belonging to each set category respectively;

[0017] Comprehensively determine the scores for the detection target belonging to each set category respectively based on the scores corresponding to each category feature.

[0018] Preferably, the category feature is the number of points feature of the target point cloud;

[0019] The category feature values include: the average number of points of the target point cloud and the upper and lower boundary values of the number of points;

[0020] The relationship curve between the category feature values of the normal target point cloud and the target distance includes: the relationship curve between the average number of points of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of the number of points of the normal target point cloud and the target distance;

[0021] The method for determining the score for the detection target belonging to the any set category includes:

[0022] Based on the relationship curve between the average number of points of the normal target point cloud and the target distance under any of the set categories, determine the average number of points corresponding to the current target distance;

[0023] Based on the relationship curve between the upper and lower boundary values of the number of points of the normal target point cloud and the target distance under any of the set categories, determine the upper and lower boundary values of the number of points corresponding to the current target distance, and calculate the fluctuation range of the number of points corresponding to the current target distance based on the upper and lower boundary values of the number of points;

[0024] Compare the current number of points of the detected target point cloud with the average number of points corresponding to the current target distance, and calculate the score corresponding to the number of points feature of the detected target belonging to any of the set categories based on the comparison result and the fluctuation range of the number of points;

[0025] Determine the score of the detected target belonging to any of the set categories based on the score corresponding to the number of points feature.

[0026] Preferably, the category feature is the SNR feature of the target point cloud;

[0027] The category feature values include: the average SNR and the upper and lower boundary values of SNR of the target point cloud;

[0028] The relationship curves between the category feature values of the normal target point cloud and the target distance include: the relationship curve between the average SNR of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of SNR of the normal target point cloud and the target distance;

[0029] The method for determining the score of the detected target belonging to any of the set categories includes:

[0030] Based on the relationship curve between the average SNR of the normal target point cloud and the target distance under any of the set categories, determine the average SNR corresponding to the current target distance;

[0031] Based on the relationship curve between the upper and lower boundary values of SNR of the normal target point cloud and the target distance under any of the set categories, determine the upper and lower boundary values of SNR corresponding to the current target distance, and calculate the SNR fluctuation range corresponding to the current target distance based on the upper and lower boundary values of SNR;

[0032] Compare the current average SNR of the detected target point cloud with the average SNR corresponding to the current target distance, and calculate the score corresponding to the SNR feature of the detected target belonging to any of the set categories based on the comparison result and the SNR fluctuation range;

[0033] Determine the score of the detected target belonging to any of the set categories based on the score corresponding to the SNR feature.

[0034] Preferably, for any one of the setting categories, the method for determining whether the detection target belongs to the any one of the setting categories based on the score indicating that the detection target belongs to the any one of the setting categories includes:

[0035] If the score indicating that the detection target belongs to the any one of the setting categories is less than the score threshold corresponding to the any one of the setting categories, it is determined that the detection target does not belong to the any one of the setting categories; otherwise, it is determined that the detection target belongs to the any one of the setting categories;

[0036] Or,

[0037] If the score indicating that the detection target belongs to the any one of the setting categories is less than the score threshold corresponding to the any one of the setting categories, the confidence level indicating that the detection target belongs to the any one of the setting categories is reduced. After the confidence level is lower than the confidence level threshold corresponding to the any one of the setting categories, it is determined that the detection target does not belong to the any one of the setting categories.

[0038] Preferably, when pre-statistically analyzing the relationship between the category features of normal targets and the target distance in each setting category of the current scenario, the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition is recorded. If the ratio of any radar target trajectory corresponding to the same normal target trajectory within a set time reaches the set binding threshold, the radar target trajectory and the same normal target trajectory are bound, and the relationship between the target distance and the category features is statistically analyzed using the bound radar target trajectory and normal target trajectory; or,

[0039] When pre-statistically analyzing the relationship between the category features of normal targets and the target distance in each setting category of the current scenario, the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition is recorded. If the corresponding relationship of any pair of radar target trajectory and normal target trajectory meets the preset conditions, the binding confidence level maintained for the any pair of radar target trajectory and normal target trajectory is increased; otherwise, the binding confidence level is reduced; when the binding confidence level is greater than or equal to the binding confidence level threshold, the any pair of radar target trajectory and normal target trajectory is bound, and the relationship between the target distance and the category features is statistically analyzed using the bound radar target trajectory and normal target trajectory; wherein, the preset conditions are: the any pair of radar target trajectory and normal target trajectory is marked as corresponding in the corresponding relationship determined based on the RV calibration result, or the ratio of the number of times the any pair of radar target trajectory and normal target trajectory is marked as corresponding within a set time is greater than the binding threshold.

[0040] An interference filtering device for millimeter-wave radar detection includes: a data acquisition unit, a scoring unit, and an interference identification unit;

[0041] The data acquisition unit is configured to obtain the point cloud data of the detection target through a millimeter-wave radar, and determine the current category feature and the current target distance of the point cloud based on the point cloud data;

[0042] The scoring unit is configured to, based on the current category feature and the current target distance, compare with the relationship between the category feature and the target distance of normal targets under each set category of the current scene determined in advance, and determine the scores of the detection target belonging to each set category;

[0043] The interference recognition unit is configured to determine whether the detection target belongs to each set category based on the scores. If the detection target does not belong to any set category, it is determined that the detection target is an interference target; it is also configured to filter the point cloud data of the interference target.

[0044] Preferably, the device further includes a statistical unit configured to determine the relationship between the category feature and the target distance of normal targets under each set category of the current scene in advance;

[0045] Among them, the determination method of the relationship between the category feature and the target distance of normal targets under any set category feature of the current scene includes:

[0046] Statistically analyze the category feature values of the target point clouds of multiple normal targets under the any set category at multiple target distances, and fit to obtain the relationship curve between the category feature value of the normal target point cloud and the target distance under the any set category.

[0047] Preferably, in the scoring unit, when there are multiple category features, the determination of the scores of the detection target belonging to each set category includes:

[0048] For any category feature, based on the relationship between the any category feature and the target distance of normal targets under each set category, determine the scores corresponding to the any category feature of the detection target belonging to each set category;

[0049] Comprehensively determine the scores of the detection target belonging to each set category based on the scores corresponding to each category feature.

[0050] Preferably, the category feature is the point number feature of the target point cloud;

[0051] The category feature values include: the average value of the point numbers of the target point cloud and the upper and lower boundary values of the point numbers;

[0052] The relationship curve between the category feature value of the normal target point cloud and the target distance includes: the relationship curve between the average value of the point numbers of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of the point numbers of the normal target point cloud and the target distance;

[0053] In the scoring unit, the method for determining the score of the detection target belonging to any of the set categories includes:

[0054] Based on the relationship curve between the average number of points of the normal target point cloud and the target distance under any of the set categories, determine the average number of points corresponding to the current target distance;

[0055] Based on the relationship curve between the upper and lower boundary values of the number of points of the normal target point cloud and the target distance under any of the set categories, determine the upper and lower boundary values of the number of points corresponding to the current target distance, and calculate the fluctuation range of the number of points corresponding to the current target distance based on the upper and lower boundary values of the number of points;

[0056] Compare the current number of points of the detection target point cloud with the average number of points corresponding to the current target distance, and calculate the point score of the detection target belonging to any of the set categories based on the comparison result and the fluctuation range of the number of points;

[0057] Determine the score of the detection target belonging to any of the set categories based on the point score.

[0058] Preferably, the category feature is the SNR feature of the target point cloud;

[0059] The category feature values include: the average SNR and the upper and lower boundary values of SNR of the target point cloud;

[0060] The relationship curves between the category feature values of the normal target point cloud and the target distance include: the relationship curve between the average SNR of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of SNR of the normal target point cloud and the target distance;

[0061] The method for determining the score of the detection target belonging to any of the set categories includes:

[0062] Based on the relationship curve between the average SNR of the normal target point cloud and the target distance under any of the set categories, determine the average SNR corresponding to the current target distance;

[0063] Based on the relationship curve between the upper and lower boundary values of SNR of the normal target point cloud and the target distance under any of the set categories, determine the upper and lower boundary values of SNR corresponding to the current target distance, and calculate the SNR fluctuation range corresponding to the current target distance based on the upper and lower boundary values of SNR;

[0064] Compare the current average SNR of the detection target point cloud with the average SNR corresponding to the current target distance, and calculate the SNR score of the detection target belonging to any of the set categories based on the comparison result and the SNR fluctuation range;

[0065] Determine the score indicating that the detection target belongs to any of the set categories based on the SNR score.

[0066] Preferably, in the interference recognition unit, for any one of the set categories, the method for determining whether the detection target belongs to the any one of the set categories based on the score indicating that the detection target belongs to the any one of the set categories includes:

[0067] If the score indicating that the detection target belongs to the any one of the set categories is less than the score threshold corresponding to the any one of the set categories, determine that the detection target does not belong to the any one of the set categories; otherwise, determine that the detection target belongs to the any one of the set categories;

[0068] Or,

[0069] If the score indicating that the detection target belongs to the any one of the set categories is less than the score threshold corresponding to the any one of the set categories, reduce the confidence level indicating that the detection target belongs to the any one of the set categories. After the confidence level is lower than the confidence level threshold corresponding to the any one of the set categories, determine that the detection target does not belong to the any one of the set categories.

[0070] Preferably, when the statistical unit pre-statistics the relationship between the category characteristics of normal targets and the target distance in each set category of the current scene, record the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition. If the ratio of any radar target trajectory corresponding to the same normal target trajectory reaches the set binding threshold within the set time, bind the radar target trajectory and the same normal target trajectory, and use the bound radar target trajectory and normal target trajectory to perform the statistics of the relationship between the target distance and the category characteristics; or,

[0071] When pre-statistics the relationship between the category characteristics of normal targets and the target distance in each set category of the current scene, record the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition. If the corresponding relationship of any radar target trajectory and normal target trajectory pair meets the preset conditions, increase the binding confidence level maintained for the any radar target trajectory and normal target trajectory pair; otherwise, reduce the binding confidence level; when the binding confidence level is greater than or equal to the binding confidence level threshold, bind the any radar target trajectory and normal target trajectory pair, and use the bound radar target trajectory and normal target trajectory to perform the statistics of the relationship between the target distance and the category characteristics; wherein, the preset conditions are: the any radar target trajectory and normal target trajectory pair are marked as corresponding in the corresponding relationship determined based on the RV marking result, or the ratio of the number of times the any radar target trajectory and normal target trajectory pair are marked as corresponding within the set time is greater than the binding threshold.

[0072] A computer-readable storage medium storing computer instructions thereon, and when the instructions are executed by a processor, the interference filtering method for millimeter-wave radar detection described in any one of the above can be implemented.

[0073] An electronic device, which at least includes a computer-readable storage medium and further includes a processor;

[0074] The processor is configured to read executable instructions from the computer-readable storage medium and execute the instructions to implement the interference filtering method for millimeter-wave radar detection described in any one of the above.

[0075] As can be seen from the above technical solutions, in the present application, point cloud data of a detection target is obtained through a millimeter-wave radar, and the current category feature and the current target distance of the point cloud are determined based on the point cloud data; thus, distance and category feature information for the current detection target can be obtained; next, based on the current category feature and the current target distance of the detection target, by comparing with the relationship between the category feature and the target distance of normal targets under each set category of the current scene determined in advance, the scores of the detection target belonging to each set category are determined; in this way, for each set category, the current category feature and the current target distance of the detection target are compared with the corresponding relationship of the normal target to obtain the score of the detection target belonging to the corresponding category, so as to characterize the possibility of the detection target belonging to the corresponding category. Finally, based on the scores, it is determined whether the detection target belongs to the corresponding set categories. If the detection target does not belong to any set category, it is determined that the detection target is an interference target; the point cloud data of the interference target is filtered out. Through the above processing, the relationship between the point cloud data features of the target and the target distance is fully utilized to identify interference targets, so that interference targets in millimeter-wave radar detection can be effectively identified, and the accuracy of millimeter-wave radar detection can be improved. Description of the Drawings

[0076] Figure 1a It is a schematic diagram of the positions where the radar target and the point cloud are projected onto the ground;

[0077] Figure 1b It is a schematic diagram of the target track and the point cloud position in the radar coordinate system;

[0078] Figure 2a It is a schematic diagram of the statistical relationship between the number of radar point clouds of large vehicles and the target distance in a traffic scene;

[0079] Figure 2b It is a schematic diagram of the statistical relationship between the number of radar point clouds of small vehicles and the target distance in a traffic scene;

[0080] Figure 3 It is a schematic diagram of the statistical relationship between the average SNR of the radar target point clouds of large vehicles and small vehicles and the target distance;

[0081] Figure 4Schematic diagram of the statistical relationship between the average SNR of the point cloud of the misdetected target FP and the normal target of the car category and the target distance;

[0082] Figure 5 Schematic diagram of the basic process of the interference filtering method for millimeter-wave radar detection in this application;

[0083] Figure 6 Schematic diagram of the specific process of the interference filtering method for millimeter-wave radar detection in the specific embodiment of this application;

[0084] Figure 7 Schematic diagram of the relationship curve between the average value of the CFAR points of the normal target of the car category and the upper and lower boundary values and the target distance;

[0085] Figure 8 Schematic diagram of the basic structure of the interference filtering device in this application;

[0086] Figure 9 Schematic diagram of the basic structure of the electronic device in this application. Detailed implementation mode

[0087] In order to make the purpose, technical means, and advantages of this application clearer, the following further elaborates on this application in conjunction with the accompanying drawings.

[0088] In the current scenario of using millimeter-wave radar for traffic monitoring, the radar track information obtained by the millimeter-wave radar is usually used. In fact, the original radar point cloud information can also be obtained through the radar. For example, it can be the coordinates, speed, signal-to-noise ratio (SNR), radar cross-section (RCS), etc. of each point in the point cloud. The characteristic information of the entire point cloud can be obtained from the information of these points.

[0089] The applicant found through research on the radar point cloud information that different types of targets can be distinguished based on the distribution information of the point cloud characteristics at different target distances. At the same time, misdetected targets and normal targets can also be distinguished.

[0090] Specifically, the distribution information of the point cloud characteristics at different target distances, that is, the relationship between the point cloud characteristics and different target distances. Taking the relationship between the number of points in the target point cloud and the average SNR and the target distance as an example, Figure 2a is the statistical relationship between the number of radar point clouds of large vehicles and the target distance in the traffic scenario, Figure 2b is the statistical relationship between the number of radar point clouds of small vehicles and the target distance in the traffic scenario. Among them, the abscissa represents the target distance, and the ordinate represents the constant false alarm rate (CFAR) points of the target point cloud. The point range of multiple same-type targets statistically represented by the bar line at each target distance, and the point on the bar line represents the statistical mean value of the points of multiple same-type targets. From Figure 2a and Figure 2bAs can be seen from the comparison, the large vehicle type and the small vehicle type can be effectively distinguished by the number of CFAR points at various target distances. Similarly, Figure 3 is the statistical relationship between the average SNR of the radar target point clouds of large vehicles and small vehicles and the target distance. An intuitive comparison of the two different types of targets is given in this figure. Among them, the upper bar lines and points represent the SNR range of large vehicles and the relationship between the average SNR and the target distance, and the lower bar lines and points represent the SNR range of small vehicles and the relationship between the average SNR and the target distance. Obviously, the SNR range interval and the average SNR of large vehicles are higher than those of small vehicles. The large vehicle type and the small vehicle type can also be effectively distinguished by the average SNR at various target distances. The above situation shows the possibility of classifying different types of targets based on the statistical values of these point cloud features at different target distances.

[0091] Through a series of statistics by the applicant, it is found that for other typical point cloud features, such as velocity consistency and RCS, etc., different types of targets can also be effectively distinguished. As can be seen from the above, different types of normal targets can be effectively distinguished by the relationship between these point cloud features and the target distance. In this application, these point cloud features that can realize the distinction of different types of targets are called the category features of the point cloud. The normal targets here refer to the actual targets in radar detection, rather than the misdetection targets. Similarly, through the relationship between the category features of the point cloud and the target distance, various types of normal targets and misdetection targets can also be distinguished. Still taking the relationship between the average SNR and the target distance as an example, Figure 4 shows the statistical relationship between the average SNR of the point cloud of the misdetection target FP and the normal target of the small vehicle category and the target distance. Among them, the connection line 1 of the average SNR and the corresponding bar lines and points correspond to the normal targets of the small vehicle category, and the connection line 2 and the corresponding bar lines and points correspond to the misdetection targets. It can be seen that there are obvious differences in the trend of the average SNR values of the multipath misdetection targets and normal vehicles changing with distance. The SNR value of the multipath misdetection target generally shows an increasing trend as the distance increases, while the SNR value of the normal vehicle target shows a decreasing trend. This shows the feasibility of distinguishing normal targets from noise targets. Through the relationship between the average SNR and the target distance, misdetection targets and normal targets can be distinguished. And through experimental statistics, it is found that for other point cloud category features, misdetection targets and normal targets can be distinguished through their relationship with the target distance.

[0092] Based on the above analysis, this application proposes to use the relationship between the point cloud category features and the target distance to identify misdetection targets. Figure 5 is a schematic diagram of the basic process of the interference filtering method for millimeter-wave radar detection in this application. As Figure 5 shown, this method includes:

[0093] Step 501, obtain the point cloud data of the detection target through a millimeter-wave radar.

[0094] Here, the detection target refers to the target newly detected by the millimeter-wave radar, and this method is used to determine whether the detection target is an interference target. Specifically, the acquisition of point cloud data can be carried out in units of frames. For example, the entire method can be applied to the single-frame point cloud data of the detection target.

[0095] Step 502: Determine the current category feature and the current target distance of the point cloud based on the point cloud data.

[0096] Analyze the point cloud data of the detection target to determine its category feature and target distance. To distinguish them from the category features and target distances of subsequent normal targets, the category feature and target distance determined based on the point cloud data of the detection target are respectively referred to as the current category feature and the current target distance. Among them, the category feature can be preset. For example, it can be the number of points of the target point cloud, the SNR of the target point cloud, the velocity consistency of the target point cloud, or the RCS of the target point cloud, etc. The number of points of the target point cloud here can be the number of points of a certain type of point cloud generated after the preprocessing of the radar raw data, such as the number of CFAR points, etc. Among them, there can be multiple different algorithms for CFAR detection, such as CA-CFAR, GO-CFAR, SO-CFAR, OS-CFAR, etc.

[0097] Step 503: Based on the current category feature and the current target distance, compare with the relationship between the category features and target distances of normal targets under each preset category in the current scene that has been determined in advance, and determine the scores of the detection target belonging to each preset category.

[0098] In this application, the relationship between the category features and target distances of normal targets under each preset category is determined in advance for the current scene. Among them, each preset category can refer to different classification information of normal targets in the current scene. For example, in the road traffic monitoring scene, the preset categories of normal targets can include: large vehicles, small vehicles, non-motor vehicles, etc. As mentioned above, the normal targets in this application refer to actual targets, which are distinguished from interference targets.

[0099] For each preset category, the relationship between the category feature and the target distance of normal targets under that preset category is determined in advance. For example, the relationship between the SNR feature and the target distance of normal targets of the large vehicle type is determined in advance, and the relationship between the SNR feature and the target distance of normal targets of the small vehicle type is determined. At the same time, the current category feature and the current target distance of the detection target are determined through the aforementioned step 502. In this step, for each preset category, referring to the relationship between the category feature of normal targets under that preset category and the current target distance, and combining the current category feature and the current target distance of the detection target, the score of the detection target belonging to the corresponding preset category is determined, which is also called the score corresponding to the corresponding preset category.

[0100] Additionally, optionally, when there are multiple category features, for example, both SNR features and point count features, for each category feature, the relationship between the normal target's category feature and the target distance can be used to determine the score corresponding to the category feature belonging to each set category. For example, determine the scores corresponding to the SNR features and point count features belonging to each set category, and then combine the scores corresponding to each category feature to determine the final score corresponding to each set category.

[0101] Step 504, based on the scores in Step 503, determine whether the detected target belongs to each set category. If the detected target does not belong to any set category, then determine that the detected target is an interference target.

[0102] For each set category, based on the score corresponding to the set category, determine whether the detected target belongs to the corresponding set category; for example, through Step 503, determine the score of the detected target belonging to the large vehicle type, and in this step, determine whether the detected target belongs to the large vehicle type according to this score. The above judgment is performed for each set category. If the judgment result is that the detected target does not belong to any set category, that is to say, the current category feature of the detected target does not conform to the relationship law between the category feature of the normal target and the target distance under any set category, then the detected target does not belong to any set category, and it is determined that it belongs to the interference target.

[0103] Step 505, filter out the point cloud data of the interference target.

[0104] For the determined interference target, filter out its point cloud data to remove the interference data, thereby effectively improving the accuracy of millimeter-wave radar detection.

[0105] So far, the interference filtering method process in this application ends. Through the above method, interference targets in millimeter-wave radar can be effectively identified, and the target detection accuracy of millimeter-wave radar can be improved.

[0106] The following uses specific embodiments to illustrate the specific implementation of this application. Figure 6 It is a schematic diagram of the specific process of the interference filtering method for millimeter-wave radar detection in a specific embodiment of this application. Among them, taking the tunnel scene and the category features of SNR and CFAR points as examples for illustration, as Figure 6 shown, this method includes:

[0107] Step 601, statistically determine the relationship between the point cloud SNR and CFAR points of normal targets and the target distance in each set category of the current scene.

[0108] In this embodiment, for the sake of simplicity, it is assumed that the set categories include large vehicles and small vehicles. For the categories of large vehicles and small vehicles, the specific statistical method for determining the relationship between the category features and the target distance is the same. This embodiment mainly considers the CFAR points and SNR features of radar targets and statistically analyzes their distributions at different distances. Among them, the category feature of CFAR points can be reflected by the mean number of points of the target point cloud and / or the upper and lower boundary values of the number of points, which is called the category feature value; the category feature value of SNR can include the average SNR of the target point cloud and / or the upper and lower boundary values of SNR. Specifically in this embodiment, the relationships between the mean number of points of the target point cloud, the upper and lower boundary values of the number of points, the average SNR, and the upper and lower boundary values of SNR and the target distance are statistically analyzed.

[0109] The statistical method can adopt various existing methods. Here, an exemplary statistical method is given. The radar-camera data is labeled synchronously, that is, the point cloud data in each frame collected by the radar is corresponding to the image data collected by the camera. The set category of the target individual (that is, the target is a large vehicle or a small vehicle) is marked from the image data collected by the camera, and the point cloud category feature value and point cloud distance of the corresponding target individual are recorded from the radar point cloud data, and the value of the point cloud category feature value - point cloud distance - target type of this target individual is recorded. Note that in order to remove some extreme values, in this embodiment, the values of each feature value in the middle 20%-80% range of the size arrangement in each distance range are taken. Therefore, the upper and lower boundary values of CFAR points and SNR in the category feature value are not extreme values, but the boundary values in the range of 20%-80%. Statistical analysis is carried out using the data marked with multiple normal targets at various target distances, and the relationship curve between the category feature value and the target distance is obtained by fitting. In this embodiment, three curves obtained by linear regression fitting of the statistical results of the average value and the upper and lower boundary values of CFAR points of multiple small vehicle targets in a certain tunnel scenario and the target distance are as Figure 7As shown, from top to bottom are the relationship curves of the upper boundary value of the CFAR points of the car with the target distance, the relationship curve of the average value of the CFAR points of the car with the target distance, and the relationship curve of the lower boundary value of the CFAR points of the car with the target distance; meanwhile, for the statistical results of the average value, upper and lower boundary values of the CFAR points of multiple large vehicle targets with the target distance, the three curves obtained by linear regression fitting are respectively the relationship curve of the upper boundary value of the CFAR points of the large vehicle with the target distance, the relationship curve of the average value of the CFAR points of the large vehicle with the target distance, and the relationship curve of the lower boundary value of the CFAR points of the large vehicle with the target distance. Similarly, in the same way, three relationship curves of the large vehicle and the car with respect to SNR can also be obtained, namely the relationship curve of the upper boundary value of the SNR of the car point cloud with the target distance, the relationship curve of the average value of the SNR of the car point cloud with the target distance, the relationship curve of the lower boundary value of the SNR of the car point cloud with the target distance, the relationship curve of the upper boundary value of the SNR of the large vehicle point cloud with the target distance, the relationship curve of the average value of the SNR of the large vehicle point cloud with the target distance, and the relationship curve of the lower boundary value of the SNR of the large vehicle point cloud with the target distance. Thus, a total of 12 relationship curves are statistically obtained in this embodiment, including 3 relationship curves for each of the 2 category features under 2 set categories. Note Figure 7 The fitting curve in this is illustrated by taking a straight line as an example. In fact, it can also be a polynomial curve or an exponential curve, etc.

[0110] In addition, in the process of annotating the radar-camera data with reference to synchronization mentioned above, the coordinates in the radar coordinate system and the coordinates in the camera image coordinate system can be mutually converted and corresponded through RV calibration. Based on the RV calibration results, the corresponding relationship between the radar target trajectory and the normal target trajectory in the image acquired by the camera can be determined, and annotation can be carried out manually. However, both the labor cost and the time cost are relatively high. In the present application, when performing annotation, the corresponding relationship between the radar target trajectory determined based on the RV calibration results and the normal target trajectory determined by the camera for image acquisition can also be automatically recorded. If the ratio of any radar target trajectory corresponding to the same normal target trajectory within a set time reaches the set binding threshold, then the radar target trajectory and the same normal target trajectory are bound, and the relationship between the target distance and the category feature is statistically analyzed using the bound radar target trajectory and normal target trajectory (specifically, the corresponding radar target and normal target can be determined based on the binding relationship between the radar target trajectory and the normal target trajectory, and then the set category to which the target belongs can be determined using the normal target, and the relationship between the target distance and the category feature under the corresponding set category can be determined using the point cloud data corresponding to the radar target); alternatively, the binding confidence of each pair of radar target trajectory and normal target trajectory (usually, any radar target trajectory and any normal target trajectory form a trajectory pair) can also be maintained. If the corresponding relationship between a certain radar target trajectory and a pair of normal target trajectories meets the preset conditions, then its binding confidence is increased, otherwise its binding confidence is decreased. When the binding confidence is greater than or equal to the set binding confidence threshold, the radar target trajectory and the pair of normal target trajectories are bound; the preset conditions can be that a certain radar target trajectory and a pair of normal target trajectories are marked as corresponding in the corresponding relationship determined based on the RV calibration results, or alternatively, the ratio of the number of times a certain radar target trajectory and a pair of normal target trajectories are marked as corresponding within a set time reaches the set binding threshold. In other words, the matching pairs of relatively stable and long-term bound RV trajectories and image targets can be automatically screened, and the markers of various targets at various distances and the analysis and curve fitting processing of the corresponding data can be automatically completed using these radar trajectories and the corresponding point cloud data. Among them, RV calibration is an existing process of automatically mapping radar point cloud data to the images acquired by the camera.

[0111] Step 602: Obtain the point cloud data of the detection target through the millimeter-wave radar.

[0112] The processing of this step is the same as that of step 501 and will not be elaborated here.

[0113] Step 603: Obtain the current category feature and the current target distance of the detection target based on the point cloud data of the detection target.

[0114] For any radar detection target in the current frame, determine the distance of the target as the current target distance, and obtain the current CFAR point number of the corresponding point cloud and the current SNR mean value , as the current category feature value

[0115] Step 604: Score the detection targets for each set category

[0116] This step is used to calculate the scores of the detection targets for each set category. Taking the calculation of the score of the detection target for the large vehicle category as an example for illustration

[0117] Obtain a radar detection target and its corresponding point cloud data through Step 602. Through Step 603, the current target distance of the detection target from the radar can be obtained, and the current CFAR point number of its point cloud can be counted and the current SNR mean value (i.e., the average SNR of the detection target point cloud) . Substitute the current target distance determined in Step 603 into the two mean value curves of the relationship between the target distance and the 6 curves of the large vehicle category obtained in Step 601, and the average CFAR point number corresponding to the normal target of the large vehicle at this target distance can be obtained and the SNR mean value . Substitute the current target distance into the other 4 upper and lower boundary value curves of the relationship between the target distance and the large vehicle, and obtain the upper and lower boundary values of the CFAR point number and the SNR upper and lower boundary values corresponding to the normal target of the large vehicle at the current target distance. Taking the difference between the upper and lower boundary values of the CFAR point number can estimate the fluctuation range of the CFAR point number corresponding to the normal target of the large vehicle at the current target distance , and taking the difference between the SNR upper and lower boundary values can estimate the fluctuation range of the SNR corresponding to the normal target of the large vehicle at the current target distance .

[0118] Next, calculate the score of the detection target for the large vehicle category based on the above data. In this embodiment, the category features include the CFAR point number feature and the SNR feature. Therefore, when calculating the score, the scores corresponding to the two category features need to be obtained respectively, and then the final score is determined. Specifically, for the CFAR point number feature and the SNR feature, the scores corresponding to the CFAR point number feature (hereinafter referred to as the CFAR point number score) and the SNR feature (hereinafter referred to as the SNR score) of the detection target belonging to the large vehicle category can be obtained respectively. For the CFAR point number feature, the current point number of the detection target point cloud can be compared with the mean value of the point numbers corresponding to the current target distance , and based on the comparison result and the point number fluctuation range , calculate the point score for the detection target belonging to the large vehicle category. A more detailed scoring calculation method can be designed according to the actual scenario requirements. For the SNR feature, the current SNR mean value of the detection target point cloud and the SNR mean value corresponding to the current target distance are compared. Based on the comparison result and the SNR fluctuation range , calculate the SNR score for the detection target belonging to the large vehicle category. Finally, determine the final score for the detection target belonging to the large vehicle category according to the CFAR point score and the SNR score.

[0119] Among them, the specific calculation rules for the CFAR point score, the SNR score, and the final score can be designed according to the actual scenario requirements. In this embodiment, an exemplary method is given. First, calculate the CFAR point score of the target and the SNR score respectively. The value range of the score is from 0 to 1. The specific scoring calculation method is as follows:

[0120] ,

[0121] .

[0122] Among them, α and β are fluctuation threshold coefficients. Generally, the fluctuation threshold can be taken slightly larger than the actual range. Therefore, the values of α and β are less than 1 and greater than 0. Qualitatively speaking, the closer the current actual CFAR point or SNR of the detection target is to the mean value of the normal target, the closer its score is to 1. At most, the current true value is allowed to differ from the average value of the normal target by the fluctuation amplitude at α and β of the same distance. If it exceeds this amplitude, the score is set to 0. That is to say, the score represents the degree of conformity between the current category feature value and the mean category feature value within the fluctuation range of the normal target at the same target distance, that is, the similarity between the detection target and the normal target.

[0123] In this embodiment, the CFAR point feature and the SNR feature are used as category features. Therefore, the score for the detection target belonging to the large vehicle category finally can be determined based on the CFAR point score and the SNR score , for example, it can be . Similarly, using the statistical curve of the point cloud category feature of the small vehicle and the target distance, the score for the detection target being a small vehicle category can be calculated .

[0124] Step 605, determine whether the detection target belongs to each set category based on the scores of each set category. If it does not belong to any set category, determine that the detection target is an interference target.

[0125] When determining whether a detection target belongs to a certain set category, the following two methods can be adopted:

[0126] Method 1: If the score of the detection target belonging to a certain set category is less than the score threshold corresponding to this set category, it is determined that the detection target does not belong to this set category; otherwise, it is determined that the detection target belongs to this set category. If the detection target does not belong to any set category, it is determined that the detection target belongs to an interference target.

[0127] In this way, the single-frame point cloud data can be directly used to determine whether the detection target is an interference target, and the recognition of interference targets can meet the real-time requirements.

[0128] Method 2: Maintain a confidence level for each set category that the detection target belongs to this category. The initial value of the confidence level can be preset according to experience. If the score of the detection target belonging to a certain set category is less than the score threshold corresponding to this set category, the confidence level of the detection target belonging to this set category is reduced. After the confidence level is lower than the confidence level threshold corresponding to the set category, it is determined that the detection target does not belong to this set category. If the detection target does not belong to any set category, it is determined that the detection target belongs to an interference target.

[0129] In this way, it may be necessary to use multi-frame point cloud data to determine whether the detection target is an interference target, but the recognition of interference targets can have higher reliability.

[0130] Through tests in the actual scenario, it is found that the accuracy of interference target recognition using Method 1 can already meet the general requirements, and at the same time, it can effectively meet the real-time requirements. Therefore, in ordinary scenarios, Method 1 can be selected to determine interference targets; for scenarios with very high requirements for the reliability of interference recognition, Method 2 can be considered to determine interference targets.

[0131] In addition, during the above judgment process, if the detection target belongs to multiple set categories, the set category with the highest score is selected as the final target category. Thus, the classification of the detection target can be realized synchronously during the interference recognition process.

[0132] Step 606: Filter out the point cloud data of the interference target.

[0133] The processing of this step is the same as that of Step 505, and will not be elaborated here.

[0134] So far, Figure 6The method flow shown ends. It should be noted that in addition to the target CFAR points and SNR features in the above specific embodiments, other features such as the point cloud velocity consistency and RCS can also be used as category features for interference target recognition. These category features can be used alone for interference target recognition, or multiple category features can be used simultaneously (such as two category features used simultaneously in the specific embodiments) to classify the detection target more accurately. Through the above Figure 6 processing, the point cloud categories of various targets are automatically or manually counted and curve regression is performed in advance. In the interference recognition stage, for each newly started track and the corresponding detection target, category scoring is performed according to the above steps to determine whether it is a false detection. If it is a false detection, the point cloud data corresponding to the track is filtered out, and the filtered radar signal will no longer contain the track in the subsequent output. Thus, interference filtering in the millimeter-wave radar can be effectively realized, the accuracy of target detection can be improved, and it can run online in real time, completing target recognition and filtering at the initial frame when multi-path targets appear.

[0135] The method of the present application described above can be applied to a variety of scenarios, especially in road traffic monitoring scenarios with tunnels or high-rise buildings and obstacles on both sides.

[0136] The above is the specific implementation of the interference filtering method for millimeter-wave radar detection in the present application. The present application also provides an interference filtering device, which can be used to implement the above interference filtering method. Figure 8 It is a schematic diagram of the basic structure of the interference filtering device in the present application. As Figure 8 shown, the device includes: a data acquisition unit, a scoring unit, and an interference recognition unit.

[0137] Among them, the data acquisition unit is used to obtain the point cloud data of the detection target through the millimeter-wave radar, and determine the current category feature and the current target distance of the point cloud based on the point cloud data;

[0138] The scoring unit is used to determine the scores of the detection target belonging to each set category based on the current category feature and the current target distance, in comparison with the relationship between the category features and the target distance of normal targets under each set category in the currently determined scene;

[0139] The interference recognition unit is used to determine whether the detection target belongs to each set category based on the score. If the detection target does not belong to any set category, it is determined that the detection target is an interference target; it is also used to filter out the point cloud data of the interference target.

[0140] Optionally, the device further includes a statistics unit, which is used to determine in advance the relationship between the category features and the target distance of normal targets under each set category in the current scene;

[0141] Among them, the method for determining the relationship between the class feature of a normal target and the target distance under any set class feature of the current scene includes:

[0142] Statistically analyze the class feature values of the target point clouds of multiple normal targets under multiple target distances in any set class, and fit to obtain the relationship curve between the class feature value of the normal target point cloud and the target distance in any set class.

[0143] Optionally, the class feature is the point number feature of the target point cloud;

[0144] The class feature values include: the average point number and the upper and lower boundary values of the point number of the target point cloud;

[0145] The relationship curve between the class feature value of the normal target point cloud and the target distance includes: the relationship curve between the average point number of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of the point number of the normal target point cloud and the target distance;

[0146] In the scoring unit, the method for determining the score of the detection target belonging to any set class includes:

[0147] Based on the relationship curve between the average point number of the normal target point cloud and the target distance in any set class, determine the average point number corresponding to the current target distance ;

[0148] Based on the relationship curve between the upper and lower boundary values of the point number of the normal target point cloud and the target distance in any set class, determine the upper and lower boundary values of the point number corresponding to the current target distance, and calculate the point number fluctuation range corresponding to the current target distance based on the upper and lower boundary values of the point number ;

[0149] Compare the current point number of the detection target point cloud with the average point number corresponding to the current target distance and, based on the comparison result and the point number fluctuation range , calculate the point number score of the detection target belonging to any set class;

[0150] Based on the point number score, determine the score of the detection target belonging to any set class.

[0151] Optionally, the class feature is the SNR feature of the target point cloud;

[0152] The class feature values include: the average SNR and the upper and lower boundary values of the SNR of the target point cloud;

[0153] The relationship curves between the category feature values of normal target point clouds and the target distance include: the relationship curve between the average SNR of normal target point clouds and the target distance, and the relationship curves between the upper and lower boundary values of SNR of normal target point clouds and the target distance;

[0154] The methods for determining the score of a detection target belonging to any of the set categories include:

[0155] Based on the relationship curve between the average SNR of normal target point clouds and the target distance under any set category, determine the average SNR corresponding to the current target distance ;

[0156] Based on the relationship curves between the upper and lower boundary values of SNR of normal target point clouds and the target distance under any set category, determine the upper and lower boundary values of SNR corresponding to the current target distance, and calculate the SNR fluctuation amplitude corresponding to the current target distance based on the upper and lower boundary values of SNR ;

[0157] Compare the current average SNR of the detection target point cloud with the average SNR corresponding to the current target distance , and calculate the SNR score of the detection target belonging to any of the set categories based on the comparison result and the SNR fluctuation amplitude ;

[0158] Determine the score of the detection target belonging to any of the set categories based on the SNR score.

[0159] Optionally, in the interference recognition unit, for any one of the set categories, the method for determining whether the detection target belongs to any of the set categories based on the score of the detection target belonging to any of the set categories includes:

[0160] If the score of the detection target belonging to any of the set categories is less than the score threshold corresponding to any of the set categories, determine that the detection target does not belong to any of the set categories; otherwise, determine that the detection target belongs to any of the set categories;

[0161] Or,

[0162] If the score of the detection target belonging to any of the set categories is less than the score threshold corresponding to any of the set categories, reduce the confidence level of the detection target belonging to any of the set categories. After the confidence level is lower than the confidence level threshold corresponding to any of the set categories, determine that the detection target does not belong to any of the set categories.

[0163] Optionally, when the statistical unit pre - statistically analyzes the relationship between the category features and the target distance of normal targets in each set category of the current scene, it records the corresponding relationship between the radar target trajectory calibrated by RV and the normal target trajectory in image acquisition. If the ratio of any radar target trajectory corresponding to the same normal target trajectory reaches the set binding threshold within the set time, then bind the radar target trajectory and the same normal target trajectory, and use the bound radar target trajectory and normal target trajectory to statistically analyze the relationship between the target distance and the category features; or,

[0164] When pre - statistically analyzing the relationship between the category features and the target distance of normal targets in each set category of the current scene, record the corresponding relationship between the radar target trajectory calibrated by RV and the normal target trajectory. If the corresponding relationship of any pair of radar target trajectory and normal target trajectory meets the preset conditions, then increase the binding confidence maintained for any pair of radar target trajectory and normal target trajectory, otherwise, decrease the binding confidence; when the binding confidence is greater than or equal to the binding confidence threshold, bind any pair of radar target trajectory and normal target trajectory, and use the bound radar target trajectory and normal target trajectory to statistically analyze the relationship between the target distance and the category features; where the preset conditions are: any pair of radar target trajectory and normal target trajectory is marked as corresponding in RV marking, or the ratio of the corresponding times of any pair of radar target trajectory and normal target trajectory within the set time is greater than the binding threshold.

[0165] The present application also provides a computer - readable storage medium, which stores instructions that can execute the steps in the interference filtering method for millimeter - wave radar detection as described above when executed by a processor. In practical applications, the computer - readable medium can be included in each device / device / system of the above - mentioned embodiments, or can exist independently and not be assembled into the device / device / system. Among them, instructions are stored in the computer - readable storage medium, and the stored instructions can execute the steps in the interference filtering method for millimeter - wave radar detection as described above when executed by a processor.

[0166] According to the embodiments disclosed in the present application, the computer - readable storage medium can be a non - volatile computer - readable storage medium. For example, it can include but is not limited to: portable computer disks, hard disks, random - access memory (RAM), read - only memory (ROM), erasable programmable read - only memory (EPROM or flash memory), portable compact disk read - only memory (CD - ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above, but does not limit the scope of protection of the present application. In the embodiments disclosed in the present application, the computer - readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, device, or device.

[0167] Figure 9 An electronic device provided by this application is also described. As Figure 9 shown, it shows a schematic structural diagram of the electronic device involved in the embodiments of this application. Specifically:

[0168] The electronic device may include a processor 901 with one or more processing cores, a memory 902 with one or more computer-readable storage media, and a computer program stored in the memory and executable on the processor. When executing the program in the memory 902, the interference filtering method for millimeter-wave radar detection can be implemented.

[0169] Specifically, in actual applications, the electronic device may also include components such as a power supply 903 and an input / output unit 904. Those skilled in the art can understand that Figure 9 the structure of the electronic device shown in

[0170] does not limit the electronic device, and it may include more or fewer components than shown, or combine certain components, or have different component arrangements. Among them:

[0171] The processor 901 is the control center of the electronic device, connecting various parts of the entire electronic device through various interfaces and lines. By running or executing software programs and / or modules stored in the memory 902, and by calling data stored in the memory 902, it executes various functions of the server and processes data, thereby monitoring the entire electronic device.

[0172] The electronic device also includes a power supply 903 that powers each component, and can be logically connected to the processor 901 through a power management system, thereby implementing functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 903 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0173] The electronic device may further include an input / output unit 904. The input unit 904 can be used to receive input digital or character information, and generate a keyboard, a mouse, a joystick, and optical signal inputs related to user settings and function controls. The input unit 904 can also be used to display information input by the user or information provided to the user, as well as various graphical user interfaces, which can be composed of graphics, text, icons, videos, and any combination thereof.

[0174] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An interference filtering method for millimeter-wave radar detection, characterized in that, Including: Obtaining point cloud data of a detection target through a millimeter-wave radar; Determining the current category feature and current target distance of the point cloud based on the point cloud data; Based on the current category feature and the current target distance, comparing with the relationship between the category feature and the target distance of normal targets under each set category in the pre-determined current scene, and determining the scores of the detection target belonging to each set category; wherein, the relationship between the category feature and the target distance is the distribution information of the category feature at different target distances; when there are multiple category features, for any category feature, based on the relationship between the any category feature and the target distance of normal targets under each set category, determining the scores corresponding to the any category feature of the detection target belonging to each set category; comprehensively determining the scores of the detection target belonging to each set category based on the scores corresponding to each category feature; Judging whether the detection target belongs to each set category based on the scores. If the detection target does not belong to any set category, determining that the detection target is an interference target; Filtering the point cloud data of the interference target; The method further includes: When pre-statistically calculating the relationship between the category feature and the target distance of normal targets under each set category in the current scene, recording the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition. If the ratio of any radar target trajectory corresponding to the same normal target trajectory reaches the set binding threshold within the set time, binding the radar target trajectory and the same normal target trajectory, and using the bound radar target trajectory and normal target trajectory to perform the statistical relationship between the target distance and the category feature; or When pre-statistically calculating the relationship between the category feature and the target distance of normal targets under each set category in the current scene, recording the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition. If the corresponding relationship of any radar target trajectory and normal target trajectory pair meets the preset condition, increasing the binding confidence maintained for the any radar target trajectory and normal target trajectory pair, otherwise, decreasing the binding confidence; when the binding confidence is greater than or equal to the binding confidence threshold, binding the any radar target trajectory and normal target trajectory pair, and using the bound radar target trajectory and normal target trajectory to perform the statistical relationship between the target distance and the category feature; wherein, the preset condition is: the any radar target trajectory and normal target trajectory pair are marked as corresponding in the determined corresponding relationship based on the RV calibration result, or the ratio of the number of times the any radar target trajectory and normal target trajectory pair are marked as corresponding within the set time is greater than the binding threshold.

2. The method according to claim 1, wherein The category feature includes: the number feature of target point cloud, the signal-to-noise ratio (SNR) feature of target point cloud, the velocity consistency feature of target point cloud, and / or the radar cross section (RCS) feature.

3. The method according to claim 1, wherein The method for determining the relationship between the category feature and the target distance of normal targets under any set category in the current scene includes: Statistically analyze the class feature values of the target point clouds of multiple normal targets under multiple target distances in any of the set categories, and fit to obtain the relationship curve between the class feature value of the normal target point cloud and the target distance in any of the set categories.

4. The method according to claim 3, wherein The class feature is the point number feature of the target point cloud; The class feature values include: the average number of points and the upper and lower boundary values of the number of points of the target point cloud; The relationship curve between the class feature value of the normal target point cloud and the target distance includes: the relationship curve between the average number of points of the normal target point cloud and the target distance, and the relationship curve between the upper and lower boundary values of the number of points of the normal target point cloud and the target distance; The method for determining the score of the detection target belonging to any of the set categories includes: Based on the relationship curve between the average number of points of the normal target point cloud under any of the above - mentioned set categories and the target distance, determine the average number of points corresponding to the current target distance ; Based on the relationship curve between the upper and lower boundary values of the number of points of the normal target point cloud under any of the above-mentioned setting categories and the target distance, determine the upper and lower boundary values of the number of points corresponding to the current target distance, and calculate the amplitude of the number of points corresponding to the current target distance based on the upper and lower boundary values of the number of points ; Compare the current number of points of the detected target point cloud with the average number of points corresponding to the current target distance and calculate the score corresponding to the point feature of the detected target belonging to any of the set categories based on the comparison result and the amplitude of point number fluctuation ; Based on the score corresponding to the point number feature, determine the score of the detection target belonging to any of the set categories.

5. The method according to claim 3 or 4, characterized in that, The class feature is the SNR feature of the target point cloud; The class feature values include: the average SNR and the upper and lower boundary values of the SNR of the target point cloud; The relationship curve between the class feature value of the normal target point cloud and the target distance includes: the relationship curve between the average SNR of the normal target point cloud and the target distance, and the relationship curve between the upper and lower boundary values of the SNR of the normal target point cloud and the target distance; The method for determining the score of the detection target belonging to any of the set categories includes: Determine the SNR mean value corresponding to the current target distance based on the relationship curve between the SNR mean value of the normal target point cloud under any of the above - mentioned setting categories and the target distance ; Based on the relationship curve between the upper and lower boundary values of SNR of the normal target point cloud under any of the above-mentioned set categories and the target distance, determine the upper and lower boundary values of SNR corresponding to the current target distance, and calculate the SNR fluctuation amplitude corresponding to the current target distance based on the upper and lower boundary values of SNR ; Compare the current SNR mean of the detected target point cloud with the SNR mean corresponding to the current target distance and calculate the score corresponding to the SNR feature of the detected target belonging to any of the set categories based on the comparison result and the SNR fluctuation range ; Based on the score corresponding to the SNR feature, determine the score of the detection target belonging to any of the set categories.

6. The method according to claim 1, wherein For any of the set categories in each set category, the method for determining whether the detection target belongs to any of the set categories based on the score of the detection target belonging to any of the set categories includes: If the score of the detection target belonging to any of the set categories is less than the score threshold corresponding to any of the set categories, determine that the detection target does not belong to any of the set categories; otherwise, determine that the detection target belongs to any of the set categories; Or, If the score of the detection target belonging to any of the set categories is less than the score threshold corresponding to any of the set categories, reduce the confidence level of the detection target belonging to any of the set categories. After the confidence level is lower than the confidence level threshold of any of the set categories, determine that the detection target does not belong to any of the set categories.

7. An interference filtering device for millimeter-wave radar detection, characterized in that, Including: A data acquisition unit, a scoring unit, an interference identification unit, and a statistical unit; The data acquisition unit is used to obtain the point cloud data of the detection target through a millimeter-wave radar, and determine the current class feature and the current target distance of the point cloud based on the point cloud data; The scoring unit is used to determine the scores of the detection target belonging to each set category by comparing the relationship between the category features and the target distance of normal targets under each set category of the current scene, based on the current category features and the current target distance; wherein, the relationship between the category features and the target distance is the distribution information of the category features at different target distances; when there are multiple category features, for any one of the category features, based on the relationship between the any one of the category features and the target distance of normal targets under each set category, determine the scores corresponding to the any one of the category features of the detection target belonging to each set category; and determine the scores of the detection target belonging to each set category by integrating the scores corresponding to each category feature. The interference recognition unit is used to determine whether the detection target belongs to each set category based on the scores. If the detection target does not belong to any set category, it is determined that the detection target is an interference target; and it is also used to filter the point cloud data of the interference target. The statistics unit is used to pre-determine the relationship between the category features and the target distance of normal targets under each set category of the current scene; wherein, when pre-statistically analyzing the relationship between the category features and the target distance of normal targets under each set category of the current scene, record the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition. If the ratio of any radar target trajectory corresponding to the same normal target trajectory within a set time reaches the set binding threshold, then bind the radar target trajectory and the same normal target trajectory, and use the bound radar target trajectory and normal target trajectory to conduct statistical analysis on the relationship between the target distance and the category features; or, when pre-statistically analyzing the relationship between the category features and the target distance of normal targets under each set category of the current scene, record the corresponding relationship between the radar target trajectory determined based on the RV calibration result and the normal target trajectory in the image acquisition. If the corresponding relationship of any radar target trajectory and normal target trajectory pair meets the preset conditions, then increase the binding confidence maintained for the any radar target trajectory and normal target trajectory pair, otherwise, decrease the binding confidence; when the binding confidence is greater than or equal to the binding confidence threshold, bind the any radar target trajectory and normal target trajectory pair, and use the bound radar target trajectory and normal target trajectory to conduct statistical analysis on the relationship between the target distance and the category features; wherein, the preset conditions are: the any radar target trajectory and normal target trajectory pair are marked as corresponding in the corresponding relationship determined based on the RV marking result, or, the ratio of the number of times the any radar target trajectory and normal target trajectory pair are marked as corresponding within a set time is greater than the binding threshold.

8. The device according to claim 7, wherein In the statistics unit, the determination method of the relationship between the category features and the target distance of normal targets under any set category feature of the current scene includes: Statistically analyze the category feature values of the target point clouds of multiple normal targets under any set category at multiple target distances, and fit to obtain the relationship curve between the category feature values of the normal target point clouds and the target distance under any set category.

9. The device according to claim 8, characterized in that, The class feature is the point number feature of the target point cloud; The class feature values include: the average point number of the target point cloud and the upper and lower boundary values of the point number; The relationship curves between the class feature values of the normal target point cloud and the target distance include: the relationship curve between the average point number of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of the point number of the normal target point cloud and the target distance; In the scoring unit, the method for determining the score of the detection target belonging to any of the set classes includes: Based on the relationship curve between the average number of points of the normal target point cloud under any of the above-mentioned set categories and the target distance, determine the average number of points corresponding to the current target distance ; Based on the relationship curve between the upper and lower boundary values of the number of points of the normal target point cloud under any of the above - mentioned setting categories and the target distance, determine the upper and lower boundary values of the number of points corresponding to the current target distance, and calculate the fluctuation amplitude of the number of points corresponding to the current target distance based on the upper and lower boundary values of the number of points ; Compare the current number of points of the detected target point cloud with the average number of points corresponding to the current target distance and calculate the point score of the detected target belonging to any of the set categories based on the comparison result and the amplitude of the point fluctuation ; Determining the score of the detection target belonging to any of the set classes based on the point number score.

10. The device according to claim 8 or 9, characterized in that, The class feature is the SNR feature of the target point cloud; The class feature values include: the average SNR of the target point cloud and the upper and lower boundary values of the SNR; The relationship curves between the class feature values of the normal target point cloud and the target distance include: the relationship curve between the average SNR of the normal target point cloud and the target distance and the relationship curve between the upper and lower boundary values of the SNR of the normal target point cloud and the target distance; The method for determining the score of the detection target belonging to any of the set classes includes: Determine the SNR mean value corresponding to the current target distance based on the relationship curve between the SNR mean value of the normal target point cloud under any of the above - mentioned set categories and the target distance ; Based on the relationship curves of the upper and lower boundary values of the SNR of the normal target point cloud under any of the set categories with the target distance, determine the upper boundary value and the lower boundary value of the SNR corresponding to the current target distance, and calculate the SNR fluctuation amplitude corresponding to the current target distance based on the upper boundary value and the lower boundary value of the SNR ; Compare the current SNR mean of the detected target point cloud with the SNR mean corresponding to the current target distance and calculate the SNR score of the detected target belonging to any of the set categories based on the comparison result and the SNR fluctuation range . Determining the score of the detection target belonging to any of the set classes based on the SNR score.

11. The device according to claim 7, wherein In the interference recognition unit, for any of the set classes in each set class, the method for determining whether the detection target belongs to any of the set classes based on the score of the detection target belonging to any of the set classes includes: If the score of the detection target belonging to any of the set classes is less than the score threshold corresponding to any of the set classes, it is determined that the detection target does not belong to any of the set classes; otherwise, it is determined that the detection target belongs to any of the set classes; Or, If the score of the detection target belonging to any of the set classes is less than the score threshold corresponding to any of the set classes, the confidence level of the detection target belonging to any of the set classes is reduced. After the confidence level is lower than the confidence level threshold of any of the set classes, it is determined that the detection target does not belong to any of the set classes.

12. A computer-readable storage medium having computer instructions stored thereon, characterized in that, When the instruction is executed by the processor, it can implement the interference filtering method for millimeter-wave radar detection described in any one of claims 1 to 6.

13. An electronic device, characterized in that, The electronic device includes at least a computer-readable storage medium and also includes a processor; The processor is configured to read an executable instruction from the computer-readable storage medium and execute the instruction to implement the interference filtering method for millimeter-wave radar detection described in any one of claims 1 to 6 above.

Images