Target detection device and target detection program

Abstract

To provide target detection technology for clustering the signal point cloud of a distance measurement sensor, in which targets having different speeds are clustered as separate targets even under a complex environment where the targets having different speeds are close to each other.SOLUTION: The present disclosure pertains to a target detection device 2 for clustering a signal point cloud in the signal point coordinates of a distance measuring sensor 1. The target detection device 2 is characterized by having: a signal point cloud coordinate division unit 21 for dividing a signal point coordinate into a plurality of speed ranges; and a clustering unit 22 for clustering a signal point cloud in the signal coordinates of each speed range as a single target or a plurality of targets and clustering a signal point cloud in signal point coordinates of different speed ranges as different targets and not as the same target.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a target detection technology that clusters a signal point cloud of a distance measurement sensor. [Background technology]

[0002] A target detection technique for clustering a signal point group of a radar sensor is disclosed in Patent Document 1 etc. As clustering techniques, k-means clustering and DBSCAN (Density-Based Spatial Clustering of Applications with Noise) are disclosed in Non-Patent Document 1 etc. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2018-059884 A [Non-patent literature]

[0004] [Non-Patent Document 1] Explanation and experiment of DBSCAN clustering, [online], S-Analysis, [Retrieved April 21, 2023], <URL:https: / / data-analysis-stats.jp / python / dbscan%E3%82%AF%E3%83%A9%E3%82%B9%E3%82%BF%E3%83%BC%E3%81%AE%E8%A7%A3%E8%AA%AC%E3%81%A8%E5%AE%9F%E9%A8%93 / > Summary of the Invention [Problem to be solved by the invention]

[0005] The problem with the clustering process of the conventional technology is shown in Figure 1. In the left column of Figure 1, the origin of the XY coordinate system is the installation position of the distance measurement sensor, the X-axis direction is the azimuth direction as seen from the distance measurement sensor, and the Y-axis direction is the depth direction as seen from the distance measurement sensor. In the first quadrant of the XY coordinate system, there are signal point groups of four stationary objects (walls, etc.) and two moving objects (people, etc.), and in the second quadrant of the XY coordinate system, there are signal point groups of three stationary objects (walls, etc.).

[0006] In the right column of Figure 1, signal points in the XY coordinate system are clustered using k-means clustering or DBSCAN. That is, the signal points of four stationary objects (walls, etc.) and two moving objects (people, etc.) in the first quadrant of the XY coordinate system are clustered by assigning "B", and the signal points of three stationary objects (walls, etc.) in the second quadrant of the XY coordinate system are clustered by assigning "A".

[0007] In this way, in a mixed environment where stationary objects and moving objects are in close proximity, the conventional technology is unable to cluster stationary objects and moving objects as separate targets. Moreover, in a mixed environment where targets having different speeds are in close proximity, the conventional technology is unable to cluster targets having different speeds as separate targets.

[0008] Therefore, in order to solve the above-mentioned problems, the present disclosure aims to provide a target detection technology that clusters signal point groups from a ranging sensor, which clusters targets having different speeds as separate targets even in a heterogeneous environment in which targets having different speeds are in close proximity to each other. [Means for solving the problem]

[0009] In order to solve the above problem, the signal point coordinates are divided into a plurality of speed ranges. Then, the signal point groups in the signal point coordinates of each speed range are clustered as a single target or a plurality of targets. Meanwhile, the signal point groups in the signal point coordinates of different speed ranges are clustered as different targets rather than as the same target.

[0010] Specifically, the present disclosure provides a target detection device that clusters signal points within a signal point coordinate of a distance measuring sensor, the target detection device comprising: a signal point coordinate division unit that divides the signal point coordinate into a plurality of speed ranges; and a clustering unit that clusters the signal point groups within the signal point coordinates of each speed range as a single target or multiple targets, and clusters the signal point groups within the signal point coordinates of different speed ranges as different targets rather than as the same target.

[0011] According to this configuration, even in a mixed environment in which targets having different speeds are close to each other, the targets having different speeds can be clustered as separate targets.

[0012] The present disclosure also provides a target detection device, wherein the signal point coordinate division unit divides the signal point coordinates into a stationary speed range and a moving speed range, and the clustering unit clusters the signal point groups within the signal point coordinates in each of the stationary speed range and the moving speed range as a single target or a plurality of targets, and clusters the signal point groups within the signal point coordinates in the different speed ranges of the stationary speed range and the moving speed range as different targets rather than as the same target.

[0013] According to this configuration, even in a mixed environment in which stationary objects and moving objects are in close proximity to each other, stationary objects and moving objects can be clustered as separate targets.

[0014] The present disclosure also relates to a target object detection device, characterized in that the signal point coordinate division unit sets speed ranges of the multiple speed regions in accordance with a speed dispersion of a detection target of the target object detection device, and divides the signal point coordinates into the multiple speed regions.

[0015] According to this configuration, when the velocity dispersion of the detection object is large / small, the velocity range of the multiple velocity regions can be set wide / narrow, so that the signal point group of the detection object can be extracted into one velocity region, and the signal point group of the detection object can be detected as one target.

[0016] The present disclosure also relates to a target detection device, characterized in that the signal point coordinate division unit sets speed ranges for the multiple speed regions so that adjacent speed regions overlap, and divides the signal point coordinates into the multiple speed regions.

[0017] According to this configuration, even if the signal point group to be detected is extracted into adjacent non-overlapping speed ranges, the signal point group to be detected can be extracted into only one of the adjacent overlapping speed ranges, and the signal point group to be detected can be detected as a single target.

[0018] The present disclosure also provides a target detection device, wherein the clustering unit includes a signal point coordinate quantization unit that quantizes the signal point coordinates into a plurality of grids within the signal point coordinates of each of the speed ranges, a signal point grid extraction unit that extracts a signal point grid in which the signal point group exists, a grid label assignment unit that assigns the same label to the signal point grids that are adjacent to each other and assigns different labels to the signal point grids that are not adjacent to each other, and a label clustering unit that clusters the signal point groups that exist in the signal point grids to which the same label has been assigned as the same target, and clusters the signal point groups that exist in the signal point grids to which the different labels have been assigned as different targets.

[0019] According to this configuration, only the number of quantization grids of the signal point coordinates is predetermined, and the number of clusters for clustering does not need to be predetermined. Furthermore, the amount of calculation for clustering depends largely only on the number of quantization grids, and does not depend largely on the number of signal point groups (which may be big data). Therefore, even if the signal point groups are distributed in a complex manner, the signal point groups can be clustered quickly and accurately.

[0020] The present disclosure also provides a target detection program for causing a computer serving as the target detection device to sequentially execute the processing steps executed by each processing unit included in the target detection device described above.

[0021] According to this configuration, it is possible to provide a program having the above-described effects.

[0022] The above disclosed inventions can be combined to the greatest extent possible. Effect of the Invention

[0023] In this way, the present disclosure provides a target detection technology that clusters signal point clouds from a ranging sensor, which can cluster targets having different speeds as separate targets even in a chaotic environment in which targets having different speeds are in close proximity. [Brief description of the drawings]

[0024] [Figure 1] FIG. 1 illustrates a problem with the clustering process of the prior art. [Diagram 2] FIG. 1 is a diagram illustrating an overview of the clustering process of the present disclosure. [Diagram 3] 1 is a diagram showing a configuration of a target detection system according to the present disclosure. [Figure 4] FIG. 11 is a diagram showing the procedure of clustering processing (coordinates of all signal points) according to the present disclosure. [Diagram 5] FIG. 13 is a diagram showing a specific example of the clustering process (all signal point coordinates) according to the present disclosure. [Figure 6] FIG. 13 is a diagram showing a specific example of the clustering process (all signal point coordinates) according to the present disclosure. [Figure 7] FIG. 13 is a diagram showing a specific example of the signal point coordinate division process of the present disclosure. [Figure 8] FIG. 13 is a diagram showing a specific example of the signal point coordinate division process of the present disclosure. [Figure 9] FIG. 13 is a diagram showing a specific example of the signal point coordinate division process of the present disclosure. [Figure 10]FIG. 1 is a diagram showing the procedure of the clustering process (for each speed range) according to the present disclosure. [Figure 11] FIG. 13 is a diagram showing a specific example of the clustering process (for each speed range) according to the present disclosure. [Figure 12] FIG. 13 is a diagram showing a specific example of the clustering process (for each speed range) according to the present disclosure. [Figure 13] FIG. 4 is a diagram showing a procedure of target oversight suppression processing according to the present disclosure. [Figure 14] FIG. 11 is a diagram showing a specific example of the target oversight suppression process of the present disclosure. [Figure 15] FIG. 11 is a diagram showing a specific example of the target oversight suppression process of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0025] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present disclosure will be described with reference to the accompanying drawings. The embodiments described below are examples of the present disclosure, and the present disclosure is not limited to the following embodiments.

[0026] (Overview of the clustering process of the present disclosure) An overview of the clustering process of the present disclosure is shown in Fig. 2. In the left column of Fig. 2, the origin of the XY coordinate system is the installation position of the distance measurement sensor, the X-axis direction is the azimuth direction as seen from the distance measurement sensor, and the Y-axis direction is the depth direction as seen from the distance measurement sensor. In the first quadrant of the XY coordinate system, there are signal point groups of four stationary objects (walls, etc.) and two moving objects (people, etc.), and in the second quadrant of the XY coordinate system, there are signal point groups of three stationary objects (walls, etc.).

[0027] In the right column of Fig. 2, k-means clustering, DBSCAN, or the clustering of the present disclosure is used to cluster the signal points in the XY coordinates of each speed range V. That is, the signal points of four stationary objects (walls, etc.) existing in the first quadrant of the XY coordinates of the stationary speed range V are clustered by assigning "B", the signal points of three stationary objects (walls, etc.) existing in the second quadrant of the XY coordinates of the stationary speed range V are clustered by assigning "A", and the signal points of two moving objects (people, etc.) existing in the first quadrant of the XY coordinates of the moving speed range V are clustered by assigning "C".

[0028] In this manner, the present disclosure allows for clustering of stationary and moving objects as separate targets even in a mixed environment in which stationary and moving objects are in close proximity, and moreover, the present disclosure allows for clustering of targets having different speeds as separate targets even in a mixed environment in which targets having different speeds are in close proximity.

[0029] The configuration of the target detection system of the present disclosure is shown in Fig. 3. The target detection system S includes a distance measurement sensor 1, a target detection device 2, and a target display device 3. The target detection device 2 includes a signal point coordinate division unit 21 and a clustering unit 22. The clustering unit 22 includes a signal point coordinate quantization unit 23, a signal point grid extraction unit 24, a grid label assignment unit 25, and a label clustering unit 26. The target detection device 2 can be realized by installing a target detection program shown in Figs. 4, 10, and 13, which will be described later, in a computer.

[0030] The distance measurement sensor 1 is a radar sensor, a LiDAR (Light Detection And Ranging) sensor, a TOF (Time Of Flight) camera, a sonar sensor, etc. The target detection device 2 clusters a group of signal points within the signal point coordinates (within two-dimensional or three-dimensional coordinates) of the distance measurement sensor 1. The target display device 3 displays the target detection results (position, movement, speed, etc. of the target) of the target detection device 2 on a display screen.

[0031] The target detection device 2 uses k-means clustering or DBSCAN to cluster the signal points in the signal point coordinates of each speed range, as shown in Figures 4 to 9. Alternatively, the target detection device 2 uses the clustering of the present disclosure to cluster the signal points in the signal point coordinates of each speed range, as shown in Figures 10 to 15.

[0032] (Procedure of clustering process (coordinates of all signal points) according to the present disclosure) The procedure of the clustering process (all signal point coordinates) of the present disclosure is shown in Fig. 4. Specific examples of the clustering process (all signal point coordinates) of the present disclosure are shown in Fig. 5 and Fig. 6. The signal point coordinate division unit 21 inputs data of a group of signal points in the signal point coordinates of the ranging sensor 1 from the ranging sensor 1. In Fig. 5 and Fig. 6, in the total signal point coordinate data D0, the origin of the XY coordinates is the installation position of the ranging sensor 1, the X-axis direction is the azimuth direction as seen from the ranging sensor 1, the Y-axis direction is the depth direction as seen from the ranging sensor 1, and the V-axis direction is the relative speed as seen from the ranging sensor 1.

[0033] The signal point coordinate division unit 21 divides the signal point coordinates into a plurality of speed ranges (step S1). In Fig. 5, the entire signal point coordinate data D0 is divided into signal point group data D1-1 of a positive high speed range, signal point group data D1-2 of a positive low speed range, signal point group data D1-3 of a stationary speed range, signal point group data D1-4 of a negative low speed range, and signal point group data D1-5 of a negative high speed range.

[0034] The signal point coordinate division unit 21 may divide the signal point coordinates into a stationary speed region and a moving speed region (step S1). In Fig. 6, all signal point coordinate data D0 is divided into signal point group data D1-1 of a positive moving speed region (high speed region and low speed region), signal point group data D1-2 of a stationary speed region, and signal point group data D1-3 of a negative moving speed region (high speed region and low speed region).

[0035] The clustering unit 22 clusters the signal points in the signal point coordinates of each speed range as a single target or multiple targets, and clusters the signal points in the signal point coordinates of different speed ranges as different targets rather than the same target (step S2). In Fig. 5, two signal points are clustered as single targets "D", "B", "A", "C", and "E" in the signal point group data D1-1, D1-2, D1-3, D1-4, and D1-5, respectively, using k-means clustering or DBSCAN.

[0036] In this way, even in a cluttered environment in which targets having different velocities are in close proximity to each other, targets having different velocities can be clustered as separate targets.

[0037] The clustering unit 22 may cluster the signal points in the signal point coordinates of each of the stationary speed range and the moving speed range as a single target or multiple targets, and may cluster the signal points in the signal point coordinates of different speed ranges of the stationary speed range and the moving speed range as different targets rather than the same target (step S2). In Fig. 6, two signal points are clustered as single targets "B", "A", and "C" in the signal point group data D1-1, D1-2, and D1-3, respectively, using k-means clustering or DBSCAN. As shown in Fig. 7, when the XY coordinates of the targets "B" and "C" are almost the same, the targets "B" and "C" may be clustered as a single target.

[0038] In this way, even in a mixed environment where stationary objects and moving objects are in close proximity to each other, stationary objects and moving objects can be clustered as separate targets.

[0039] Specific examples of the signal point coordinate division process of the present disclosure are shown in Fig. 7 and Fig. 8. The signal point coordinate division unit 21 may set a plurality of speed ranges in accordance with the speed dispersion of the detection target of the target detection device 2, and divide the signal point coordinates into the plurality of speed ranges (step S1).

[0040] In FIG. 7, the detection target is a human P, and the human P walks with his / her arms and legs swinging back and forth, so the speed variance of the detection target is large. Therefore, in the total signal point coordinate data D0, the speed range of a plurality of speed regions is set widely, and the signal point group data D1-1 of a relatively large positive moving speed region, the signal point group data D1-3 of a relatively large negative moving speed region, and the signal point group data D1-2 of a stationary speed region are divided. Then, using k-means clustering or DBSCAN, seven signal point groups are clustered as a single target “A” in the signal point group data D1-1. For example, the action of the human P swinging his / her arms and legs up is distributed as two signal point groups in the positive high speed region of the target “A”, the action of the human P swinging his / her arms and legs down is distributed as two signal point groups in the positive low speed region of the target “A”, and the action of the human P moving his / her torso other than his / her arms and legs is distributed as three signal point groups in the positive medium speed region of the target “A”.

[0041] In FIG. 8, the detection target is an automobile V, which has multiple reflection points but travels as a single unit, so the speed variance of the detection target is small. Therefore, in the total signal point coordinate data D0, the speed ranges of multiple speed regions are set narrow, and the data is divided into signal point group data D1-1 of a positive high speed region, signal point group data D1-2 of a positive low speed region, signal point group data D1-3 of a stationary speed region, signal point group data D1-4 of a negative low speed region, and signal point group data D1-5 of a negative high speed region. Then, using k-means clustering or DBSCAN, four signal point groups are clustered as a single target "B" in the signal point group data D1-1. For example, the reflected signal from the body of the automobile V is clustered as the target "B". Then, the reflected signal from the tire of the automobile V is clustered as a separate target with approximately twice the speed (although not shown in FIG. 8, the rotational speed variance of the tires of the automobile V is also small).

[0042] In this way, when the velocity dispersion of the detection target is large / small, the signal point group of the detection target can be extracted into a single velocity range by setting the velocity range of multiple velocity regions to a wide / narrow range, and the signal point group of the detection target can be detected as a single target.

[0043] A specific example of the signal point coordinate division process of the present disclosure is also shown in Fig. 9. The signal point coordinate division unit 21 may set speed ranges for a plurality of speed regions so that adjacent speed regions overlap each other, and divide the signal point coordinates into a plurality of speed regions (step S1).

[0044] 9, all signal point coordinate data D0 is divided into signal point group data D1-1 of a positive moving speed region, signal point group data D1-2 of a stationary speed region, and signal point group data D1-3 of a negative moving speed region so that adjacent speed regions do not overlap. However, using k-means clustering or DBSCAN, there is a possibility that each of the signal point groups in the signal point group data D1-1 and D1-2 is not clustered as a single target (even when the XY coordinates of each signal point group are almost the same).

[0045] In the right column of Fig. 9, all signal point coordinate data D0 is divided into signal point group data D1-1' of a positive moving speed region, signal point group data D1-2' of a stationary speed region, and signal point group data D1-3' of a negative moving speed region so that adjacent speed regions overlap. Then, using k-means clustering or DBSCAN, two signal point groups (a total of two signal point groups in the left column of Fig. 9) in the signal point group data D1-1' are clustered as a single target "A" (if the XY coordinates of each signal point group are approximately the same).

[0046] In this way, even if the signal point group of the detection target is extracted into adjacent non-overlapping speed ranges, the signal point group of the detection target can be extracted into only one of the adjacent overlapping speed ranges, and the signal point group of the detection target can be detected as a single target.

[0047] In k-means clustering, signal points are clustered based on the distance of the signal points from the center of gravity of the k clusters. However, when the signal points are distributed in a complex manner (when a signal point group is distributed in a circular pattern and other signal points exist), the signal points cannot be clustered accurately. Therefore, it is necessary to determine in advance how many clusters (k) the signal points should be clustered into.

[0048] On the other hand, DBSCAN clusters signal points based on the density of the signal points, specifically, based on the distance between the signal points. Therefore, even if the signal points are distributed in a complex manner (such as when a signal point group is distributed in a circular pattern and other signal point groups exist), the signal points can be accurately clustered. However, the amount of calculation required for clustering is on the order of the square of the number of signal point groups, and increases as the number of signal point groups increases.

[0049] Therefore, the target detection device 2 uses clustering according to the present disclosure (quantization of signal point coordinates and clustering of a signal point grid) instead of k-means clustering and DBSCAN to cluster a group of signal points within the signal point coordinates of each speed range.

[0050] (Procedure of clustering process (for each speed range) according to the present disclosure) The procedure of the clustering process (for each speed range) of the present disclosure is shown in Fig. 10. Specific examples of the clustering process (for each speed range) of the present disclosure are shown in Fig. 11 and Fig. 12. The signal point coordinate quantization unit 23 inputs the signal point group data in the signal point coordinates of each speed range from the signal point coordinate division unit 21. In the upper left column of Fig. 11, in the signal point group data D1 of each speed range, the origin of the XY coordinate is the installation position of the distance measurement sensor 1, the X-axis direction is the azimuth direction as seen from the distance measurement sensor 1, and the Y-axis direction is the depth direction as seen from the distance measurement sensor 1. Then, there are signal point groups of n points (=13 points) in the XY coordinates, which can become big data.

[0051] The signal point coordinate quantization unit 23 inputs a parameter of the size of the grid and quantizes the signal point coordinates of each speed range into a plurality of grids (step S11). In the upper right column of FIG. 11, in the quantization grid data D2 of each speed range, m (=5) grids are generated in the X-axis direction and m (=5) grids are generated in the Y-axis direction, and m (=5) grids are generated in the XY coordinates. 2 (=25) grids are generated. The amount of calculation required to generate the quantized grid data D2 for all speed ranges is on the order of 1 × the number of all speed ranges. As a modified example, in the quantized grid data D2 for each speed range, m grids are generated in the X-axis direction, and n grids (which may or may not be the same number as m) are generated in the Y-axis direction, generating mn grids in the XY coordinate system. In the quantized grid data D2 for each speed range, the size of each grid may be set larger / smaller as the speed range of each speed range (see Figs. 7 and 8) becomes wider / narrower, or the size of each grid may be set arbitrarily regardless of the speed range of each speed range.

[0052] The signal point grid extraction unit 24 inputs a parameter of the minimum threshold (=2) of the signal point group, and extracts a signal point grid where a signal point group exists in each speed range (step S12). In the lower left column of Fig. 11, in the signal point count data D3 for each speed range, the grid in the pth row and qth column is set as grid (p, q), and "2" is assigned to grid (2, 1), "1" to grid (2, 4), "3" to grid (3, 4), "3" to grid (4, 3), "2" to grid (5, 1), and "2" to grid (5, 4). The amount of calculation required to generate the signal point count data D3 for all speed ranges is on the order of n (=number of signal point groups) x number of all speed ranges.

[0053] The signal point grid extraction unit 24 extracts signal point grids in which a signal point group equal to or greater than the minimum threshold (=2) exists in each speed range (steps S13, YES and step S14). On the other hand, the signal point grid extraction unit 24 discards signal point grids in which a signal point group less than the minimum threshold (=2) exists in each speed range (steps S13, NO and step S15). In the lower right column of FIG. 11 and the upper left column of FIG. 12, in the extracted grid data D4 for each speed range, the grid in the pth row and qth column is set as grid (p, q), "Y" is assigned to grid (2, 1), "N" is assigned to grid (2, 4), "Y" is assigned to grid (3, 4), "Y" is assigned to grid (4, 3), "Y" is assigned to grid (5, 1), and "Y" is assigned to grid (5, 4). The amount of calculation required to generate extracted grid data D4 for all speed ranges is m 2 (=Number of grids on each axis 2 ) × the number of all speed ranges.

[0054] When discarding a signal point grid in which a signal point group less than the minimum threshold (=2) exists, the signal point grid extraction unit 24 executes a target missed object suppression process shown in FIG.

[0055] The grid label assignment unit 25 assigns the same label to adjacent signal point grids in each speed range (step S16). On the other hand, the grid label assignment unit 25 assigns different labels to non-adjacent signal point grids in each speed range (step S16). Here, "adjacent" refers to all of the surrounding eight neighbors or only the four neighbors above, below, left and right as viewed from the signal point grid of interest. In the upper right column of FIG. 12, in the grid label data D5 for each speed range, the grid in the pth row and qth column is set as grid (p, q), "A" is assigned to grid (2, 1), "B" is assigned to grid (3, 4), "B" is assigned to grid (4, 3), "C" is assigned to grid (5, 1), and "B" is assigned to grid (5, 4). The amount of calculation required to generate the grid label data D5 for all speed ranges is m 2 (=Number of grids on each axis 2 ) × the number of all speed ranges.

[0056] The label clustering unit 26 clusters signal points present in signal point grids to which the same label has been assigned in each speed range as the same target (step S17). On the other hand, the label clustering unit 26 clusters signal points present in signal point grids to which different labels have been assigned in each speed range as different targets (step S17). 12, in the label reference data D6 for each speed range, the grid in the pth row and qth column is designated as grid (p, q), and there are two signal point groups in grid (2, 1) to which "A" is assigned, three signal point groups in grid (3, 4) to which "B" is assigned, three signal point groups in grid (4, 3) to which "B" is assigned, two signal point groups in grid (5, 1) to which "C" is assigned, two signal point groups in grid (5, 4) to which "B" is assigned, and one signal point group in grid (2, 4) to which "N" is assigned (already discarded signal point grid). The amount of calculation required to generate the label reference data D6 for all speed ranges is on the order of n (=number of signal point groups) × number of all speed ranges.

[0057] The label clustering unit 26 outputs the signal point group data clustered in each speed range to the target display device 3. In the lower right column of Fig. 12, in the clustering data D7 for each speed range, the grid in the pth row and qth column is set as grid (p, q), and "A" is assigned to the two signal point groups present in grid (2, 1) to which "A" is assigned, "B" is assigned to the eight signal point groups present in grids (3, 4), (4, 3), and (5, 4) to which "B" is assigned, "C" is assigned to the two signal point groups present in grid (5, 1) to which "C" is assigned, and one signal point group present in grid (2, 4) (already discarded signal point grid) to which "N" is assigned is discarded. The amount of calculation required to generate the clustering data D7 for all speed ranges is on the order of n (=number of signal point groups) x number of all speed ranges.

[0058] In other words, the number of quantization grids for the signal point coordinates m 2Only the number of clusters k for clustering needs to be determined in advance. The amount of calculation for clustering is reduced by the number of quantization grids m 2 The amount of calculation for clustering does not depend greatly on the number of signal points n (which can be big data). Therefore, even if the signal points are distributed in a complex manner, the signal points can be clustered quickly and accurately.

[0059] Furthermore, it is possible to extract a signal point grid where a large number (e.g., 2 or more) of signal point groups exist as a signal, and discard a signal point grid where a small number (e.g., 1 or less) of signal point groups exist as noise. However, if a target having a plurality of signal point groups is divided into adjacent grids with few signal points using the quantization grid data D2, it may be discarded as noise. Therefore, the signal point grid extraction unit 24 executes the target missed suppression process shown in FIG. 13.

[0060] (Procedure of target oversight suppression processing (for each speed range) of the present disclosure) The procedure of the target missed object suppression process (for each speed range) of the present disclosure is shown in Fig. 13. Specific examples of the target missed object suppression process (for each speed range) of the present disclosure are shown in Fig. 14 and Fig. 15. The signal point grid extraction unit 24 determines whether signal point grids (hereinafter referred to as "low signal point grids") in which signal point groups less than the minimum threshold (=2) exist in each speed range are adjacent to each other (step S21). In the left column of Fig. 14, in the signal point number data D3 for each speed range, the grid in the pth row and qth column is set as grid (p, q), and one target having two signal point groups is separated into the lower left corner of grid (2, 3) and the upper left corner of grid (3, 3), and "1" is assigned to grids (2, 3) and (3, 3). In the left column of Figure 15, in the signal point count data D3 for each speed range, the grid in the pth row and qth column is set as grid (p, q), and one target having two signal point groups is separated near the bottom edge of grid (2, 3) and near the top edge of grid (3, 3), and "1" is assigned to grids (2, 3) and (3, 3).

[0061] When the low signal point grids are adjacent to each other in each speed range (step S21, YES), the signal point grid extraction unit 24 sets a new grid that straddles the boundary between the adjacent low signal point grids (step S22). In the center column of Fig. 14, in the signal point number data D3' for each speed range, the new grid (indicated by the dashed line) is shifted by half the grid size in both the X- and Y-axis directions compared to the original grid (indicated by the solid line), the grid in the p'th row and q'th column is set as grid (p', q'), one target having two signal point groups exists inside grid (3', 3'), and "2" is assigned to grid (3', 3'). In the middle column of Figure 15, in the signal point number data D3' for each speed range, a new grid (shown by a dashed line) is shifted by half the grid size in the Y-axis direction only compared to the original grid (shown by a solid line), the grid in the p'th row and q'th column is set to grid (p', q'), one target having two signal point groups exists inside grid (3', 3'), and a "2" is assigned to grid (3', 3').

[0062] When a signal point group having a value equal to or greater than the minimum threshold (=2) exists in a new grid that straddles the boundary between low-signal-point grids in each speed range (step S23, YES), the signal point grid extraction unit 24 considers that a signal point group having a value equal to or greater than the minimum threshold (=2) exists in adjacent low-signal-point grids (step S24). In the right column of FIG. 14, the grid in the pth row and qth column in the signal point number data D3" for each speed range is designated as grid (p, q), and "2" is assigned to grids (2, 3) and (3, 3) instead of "1", and in order to more reliably execute the target missed suppression processing, "2" may be assigned to grids (2, 2) and (3, 2) instead of "0". In the right column of FIG. 15, the grid in the pth row and qth column in the signal point number data D3" for each speed range is designated as grid (p, q), and "2" is assigned to grids (2, 3) and (3, 3) instead of "1".

[0063] When a signal point group equal to or greater than the minimum threshold (=2) exists in a new grid that straddles a boundary between low signal point grids in each speed range (step S23, YES), the signal point grid extraction unit 24 extracts adjacent low signal point grids without discarding them (step S25). In the right column of FIG. 14, the grid in the pth row and qth column in the signal point number data D3" for each speed range is set to grid (p, q), and grids (2, 3) and (3, 3) are extracted without being discarded, and one target having two signal point groups is extracted without being discarded. In the right column of FIG. 15, the grid in the pth row and qth column in the signal point number data D3" for each speed range is set to grid (p, q), and grids (2, 3) and (3, 3) are extracted without being discarded, and one target having two signal point groups is extracted without being discarded.

[0064] On the other hand, when the low signal point grids are not adjacent to each other in each speed region (step S21, NO), the signal point grid extraction unit 24 discards the low signal point grid (step S26, not shown in Figs. 14 and 15).The signal point grid extraction unit 24 also discards the low signal point grid when a signal point group less than the minimum threshold (=2) exists in a new grid that straddles the boundary of the low signal point grid in each speed region (step S23, NO) (step S26, not shown in Figs. 14 and 15).

[0065] That is, when a target having a plurality of signal point groups is divided into adjacent small signal point grids using the original quantization grid, the target may be discarded as noise. However, when a target having a plurality of signal point groups is double-used, using the original quantization grid and the new quantization grid, the target may be extracted as a signal without being discarded as noise. Note that the new quantization grid may be shifted slightly from the original quantization grid not once but twice or more times, and the target missed suppression process may be performed not once but twice or more times, so that the signal may be extracted more reliably.

[0066] Note that even when both the original quantization grid and the new quantization grid are used, the amount of calculation for clustering increases by only a few times compared to when only the original quantization grid is used, and the amount of calculation for clustering does not depend greatly on the number of signal point groups n. [Industrial Applicability]

[0067] The target detection device and the target detection program disclosed herein can apply a radar sensor, a LiDAR sensor, a TOF camera, a sonar sensor, or the like as a distance measurement sensor. [Explanation of symbols]

[0068] S: Target detection system 1: Distance sensor 2: Target detection device 3:Target display device 21: Signal point coordinate division unit 22: Clustering section 23: Signal point coordinate quantization unit 24: Signal point grid extraction unit 25: Grid label assignment section 26: Label clustering part D0: All signal point coordinate data D1, D1-1, D1-2, D1-3, D1-4, D1-5, D1-1', D1-2', D1-3': Signal point group data D2: Quantized grid data D3, D3', D3”: Signal point count data D4: Extract grid data D5:Grid Label Data D6: Label Reference Data D7: Clustering data P:Human V: Automobiles

Claims

1. A target detection device that performs clustering on a signal point group within a signal point coordinate system of a distance measuring sensor, a signal point coordinate division unit that divides the signal point coordinates into a plurality of speed ranges; a clustering unit that clusters the signal points in the signal point coordinates in each speed range as a single target or a plurality of targets, and clusters the signal points in the signal point coordinates in different speed ranges as different targets rather than as the same target; A target detection device comprising:

2. The signal point coordinate division unit divides the signal point coordinates into a stationary speed region and a moving speed region, The clustering unit clusters the signal points in the signal point coordinates in each of the stationary speed range and the moving speed range as a single target or a plurality of targets, and clusters the signal points in the signal point coordinates in the different speed ranges in the stationary speed range and the moving speed range as different targets rather than the same target.

2. The target detection device according to claim 1 .

3. The signal point coordinate division unit sets speed ranges of the plurality of speed regions in accordance with a speed dispersion of the detection target of the target detection device, and divides the signal point coordinates into the plurality of speed regions.

3. The target detection device according to claim 1 or 2.

4. The signal point coordinate division unit sets speed ranges of the plurality of speed regions such that adjacent speed regions overlap each other, and divides the signal point coordinates into the plurality of speed regions.

3. The target detection device according to claim 1 or 2.

5. The clustering unit, within the signal point coordinates of each of the speed ranges, a signal point coordinate quantization unit that quantizes the signal point coordinates into a plurality of grids; a signal point grid extraction unit that extracts a signal point grid in which the signal point group exists; a grid label assignment unit that assigns the same label to adjacent signal point grids and assigns different labels to non-adjacent signal point grids; a label clustering unit that clusters the signal points present in the signal point grid to which the same label has been assigned as the same target object, and clusters the signal points present in the signal point grid to which the different labels have been assigned as the different targets; The target detection device according to claim 1 or 2, further comprising:

6. 3. A target detection program for causing a computer as the target detection device to sequentially execute each processing step executed by each processing unit included in the target detection device according to claim 1 or 2.

Images