Information processing device

Abstract

To provide an object identification device that can perform object identification more accurately. [Solution] The point cloud information is clustered into clusters, and it is checked whether the point cloud corresponding to the same object is divided into multiple clusters. If the point cloud corresponding to the same object is divided into multiple clusters, the divided clusters are merged.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus.

Background Art

[0002] Techniques for discriminating the types of vehicles traveling on a road using LiDAR (Light Detection and Ranging; Laser Imaging Detection and Ranging) have been developed (for example, Patent Document 1). In the technique disclosed in Patent Document 1, a rectangle corresponding to a cluster of point clouds is estimated, the width and / or depth of the estimated rectangle is estimated, and the vehicle corresponding to the cluster is discriminated using the estimated width and / or depth. That is, in the technique disclosed in Cited Document 1, the vehicle corresponding to the cluster is discriminated based on the size of the cluster.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Point clouds corresponding to large vehicles (for example, trucks and trailers) may be split into one or more clusters instead of being clustered into one cluster. In such a case, the size of each of the split clusters is smaller than the size of the region where the point cloud of the large vehicle exists. Therefore, when an object corresponding to each of the split clusters is discriminated based on the size of the cluster, the cluster is discriminated as an object smaller than the object (large vehicle) that the cluster actually corresponds to.

[0005] Therefore, an object of the present invention is to provide an object discrimination apparatus capable of performing more accurate object discrimination. [Means for solving the problem]

[0006] To solve the above problems, an information processing device according to one embodiment of the present invention includes: a point cloud information acquisition processing unit that acquires point cloud information of a detection area on a road; a clustering processing unit that clusters the point clouds included in the point cloud information into clusters; and a cluster merging processing unit that checks whether the point cloud corresponding to the same object is divided into multiple clusters, and if the point cloud corresponding to the same object is divided into multiple clusters, merges the divided clusters.

[0007] The cluster joining processing unit may, in each of the clusters, join a cluster that satisfies the first condition with respect to that cluster.

[0008] In each of the clusters, the first condition may be that it lies within a first length in a first direction from the representative point of the cluster.

[0009] In each of the clusters, the first condition may be such that it is located within a first length in a first direction from the representative point of the cluster and within the lane in which the object corresponding to the cluster is moving.

[0010] The cluster merging unit sets the lane region for each of the clusters based on the representative point of the cluster and the two points located at the outermost edges in the width direction of the cluster, and the width direction of the cluster may be perpendicular to the direction of travel of the cluster in the horizontal plane.

[0011] In each of the aforementioned clusters, the first direction may be the direction of progression of the cluster if the point cloud of the cluster spreads out from the representative point of the cluster in the direction of progression of the cluster, or it may be the direction opposite to the direction of progression of the cluster if the point cloud of the cluster spreads out from the representative point of the cluster in the opposite direction to the direction of progression of the cluster.

[0012] The first length may be determined based on the size of the large vehicle.

[0013] The point cloud information consists of a plurality of frames generated at a predetermined frame rate, the clustering processing unit clusters the point cloud into clusters in each frame of the point cloud information, and the cluster joining processing unit sets, in each frame of the point cloud information, clusters that satisfy a first condition with respect to each of the clusters clustered in that frame as a group of candidate clusters for joining, and if the group of candidate clusters for joining is set for a cluster, the unit may join the cluster from the group of candidate clusters that satisfies the first condition with respect to the cluster in the frame preceding that frame to that cluster.

[0014] The cluster merging processing unit may, for each of the clusters, set the first length to a first value if the height of the cluster is equal to or greater than the first height, and set the first length to a second value which is smaller than the first value if the height of the cluster is not equal to or greater than the first height.

[0015] The cluster merging unit may, for each of the clusters, merge a cluster that satisfies the first condition with the cluster if the height of the cluster is equal to or greater than the first height.

[0016] The information processing device may further include an object discrimination processing unit for each of the clusters, which determines the object corresponding to the cluster based on the size of the cluster.

[0017] An information processing method according to an embodiment of the present invention is an information processing method executed by a computer, including a point cloud information acquisition step of acquiring point cloud information of a detection area on a road, a clustering processing step of clustering the point clouds included in the point cloud information into clusters, and a cluster combining processing step of checking whether the point clouds corresponding to the same object are split into multiple clusters and combining the split clusters if so.

[0018] An information processing program according to an embodiment of the present invention causes a computer to execute the above information processing method.

Advantages of the Invention

[0019] The present invention makes it possible to provide an object discrimination device capable of more accurately discriminating objects.

Brief Description of the Drawings

[0020] [Figure 1] It is a diagram showing an object discrimination device 100 according to an embodiment of the present invention. [Figure 2] It is a diagram for explaining the relationship between the lidar 110 (object discrimination device 100), the road RW, and the detection area A. [Figure 3] It is a diagram for explaining the relationship between the lidar 110 (object discrimination device 100), the road RW, and the detection area A. [Figure 4] It is a diagram showing an example of the control unit 120. [Figure 5] It is a diagram for explaining the relationship between the detection area A and the vehicles AM1, AM2. [Figure 6] It is a diagram for explaining the point cloud information of the detection area A acquired by the lidar 110 in the case shown in FIG. 5. [Figure 7] It is a diagram for explaining the relationship between the detection area A and the large vehicle AM3. [Figure 8] It is a diagram for explaining the point cloud information of the detection area A acquired by the lidar 110 in the case shown in FIG. 7. [Figure 9]This diagram illustrates vehicle AM ​​traveling in direction DA. [Figure 10] This figure illustrates the point cloud information acquired by the lidar 110 in the case shown in Figure 9. [Figure 11] This diagram illustrates an example of a first numerical range R1 and a second numerical range R2 set for two-wheeled and four-wheeled vehicles. [Figure 12] This figure shows an example of a processing operation performed in the control unit 120. [Figure 13] This diagram illustrates the relationship between detection area A, large vehicle AM3, and driving lane LN. [Figure 14] This figure illustrates the relationship between the point cloud information acquired by the lidar 110, the first direction DS, and the first length L1 in the case shown in Figure 13. [Figure 15] This diagram illustrates the relationship between detection area A, large vehicle AM3, and driving lane LN. [Figure 16] This figure illustrates the relationship between the point cloud information acquired by the lidar 110, the first direction DS, and the first length L1 in the case shown in Figure 15. [Figure 17] This diagram illustrates the definition of the area of ​​the driving lane (LN). [Figure 18] This diagram illustrates the relationship between detection area A, the large vehicle AM3, and the small vehicle AM4. [Figure 19] This figure illustrates the point cloud information acquired by the lidar 110 in the case shown in Figure 18. [Figure 20] This diagram illustrates the relationship between detection area A, the large vehicle AM3, and the small vehicle AM4. [Figure 21] This figure illustrates the point cloud information acquired by the lidar 110 in the case shown in Figure 20. [Figure 22] Figure 12 shows an example of the processing operation performed by the cluster coupling processing unit 124 in step S1204. [Figure 23] Figure 12 shows an example of the processing operation performed by the cluster coupling processing unit 124 in step S1204. [Figure 24] Figure 12 shows an example of the processing operation performed by the cluster coupling processing unit 124 in step S1204. [Modes for carrying out the invention]

[0021] <Object discrimination device 100> Figure 1 shows an object identification device 100 according to one embodiment of the present invention. The object identification device 100 is a device that identifies objects on a road (for example, vehicles), and as shown in Figures 2 and 3, it detects objects present in a predetermined detection area A on the road RW and identifies the detected objects. In the example shown in Figure 2, the object identification device 100 is installed above the road RW and detects objects present in the three-dimensional detection area A on the road RW from above. In the example shown in Figure 3, the object identification device 100 is installed beside the road RW and detects objects present in the detection area A on the road from diagonally above. The shape of the detection area A is not limited to the shape exemplified in Figures 2 and 3 (rectangular parallelepiped) and can be set as appropriate.

[0022] The object recognition device 100 includes a LiDAR 110 and a control unit 120. The LiDAR 110 is installed near the detection area A and irradiates the detection area A with laser light. By receiving the reflected light that is reflected back by an object present in the detection area A, the LiDAR measures the distance to the object and generates three-dimensional point cloud information of the detection area A. The control unit 120 is an information processing device (computer) that processes the information. Similar to the LiDAR 110, it may be installed near the detection area A, or it may be located at a distance from the detection area A.

[0023] The rider 110 generates three-dimensional point cloud information of detection area A on the road RW at a predetermined frame rate (i.e., at a predetermined time interval). Therefore, the three-dimensional point cloud information of detection area A on the road RW consists of multiple frames generated at the predetermined frame rate.

[0024] In each frame, each point is represented in Cartesian coordinates (x, y, z). In this case, for example, as shown in Figures 2 and 3, an arbitrary position is set as the origin O, the width direction of the road RW is the x-axis direction, the direction in which the road RW extends is the y-axis direction, and the direction perpendicular to the width direction and the direction in which the road RW extends (height direction) is the z-axis direction. In the example shown in Figures 2 and 3, the position where the pole to which the rider 110 is attached is set as the origin O.

[0025] Figure 4 shows an example of the control unit 120. It includes a point cloud information acquisition processing unit 121, a clustering processing unit 122, a travel direction calculation unit 123, a cluster merging processing unit 124, and an object discrimination processing unit 125.

[0026] The point cloud information acquisition processing unit 121 acquires point cloud information of the detection area A on the road RW from the lidar 110. In this embodiment, the control unit 120 only needs to acquire information relating to the length of the object in the width direction and information relating to the length of the object in the depth direction. Therefore, the point cloud information of the detection area A acquired by the point cloud information acquisition processing unit 121 may be the three-dimensional point cloud information of the detection area A generated by the lidar 110 itself, or it may be the two-dimensional point cloud information (x-coordinate and y-coordinate values ​​of each point) from this three-dimensional point cloud information.

[0027] The clustering processing unit 122 clusters the point clouds contained in the point cloud information acquired by the point cloud information acquisition processing unit 121 into clusters in each frame of the point cloud information of the detection region A. In other words, the clustering processing unit 122 clusters the point clouds contained in each frame of the point cloud information of the detection region A and divides them into clusters.

[0028] At this time, the clustering processing unit 122 clusters the point cloud contained in each frame of the point cloud information based on the distance between two points. Specifically, the clustering processing unit 112 divides the point cloud contained in each frame so that two points with a distance of less than or equal to a first distance are included in the same cluster. The distance between two points (P1(x1, y1, z1), P2(x2, y2, z2)) is the three-dimensional distance (((x1-x2) 2 +(y1-y2) 2 +(z1-z2) 2 ) 1 / 2 ) is also acceptable, or the two-dimensional distance (((x1-x2) 2 +(y1-y2) 2 ) 1 / 2 ) is also acceptable.

[0029] Each frame of the point cloud information for detection area A may contain point clouds of multiple vehicles. As shown in Figure 5, if there are two vehicles (vehicle AM1 and vehicle AM2) within the detection area A of the road RW, the point cloud information for detection area A will contain two clusters (cluster C1 and cluster C2), as shown in Figure 6. In the examples shown in Figures 5 and 6, cluster C1 corresponds to vehicle AM1 and cluster C2 corresponds to vehicle AM2. In other words, the clustering processing unit 112 clusters the point clouds contained in each frame of the point cloud information, dividing them into clusters that correspond to objects present in detection area A.

[0030] The direction of travel calculation unit 123 calculates the direction of travel for each cluster in each frame of the point cloud information in detection area A. The direction of travel calculation unit 123 may use any method as long as it can calculate the direction of travel of the cluster. For example, the direction of travel calculation unit 123 calculates the direction of travel of a cluster using the tracking history of the cluster. In this case, the direction of travel calculation unit 123 may calculate the direction of travel of each cluster based on the movement of the representative point of the cluster. Here, for each cluster, the representative point of the cluster may be, for example, the point closest to the rider 110 among the points included in the cluster. In this case, the detection area A may be such that, for example, as shown in Figures 2 and 3, the part of the vehicle included in detection area A that is closest to the rider 110 is the front end or rear end of the vehicle.

[0031] Generally, a vehicle A traveling on a road RW travels along the direction in which the road RW extends (y-axis direction). Therefore, the cluster corresponding to vehicle A traveling on the road RW also generally moves along the direction in which the road RW extends (y-axis direction). Thus, the direction of travel calculation unit 123 may use the direction in which the road RW extends (y-axis direction) as the direction of travel of the cluster.

[0032] Each frame of the point cloud information in detection area A may contain point clouds of large vehicles (e.g., trucks or trailers). As shown in Figure 7, if a large vehicle (vehicle AM3) is present within detection area A of road RW, the point cloud information in detection area A may contain two or more clusters (cluster C3-1, cluster C3-2, cluster C3-3), as shown in Figure 8. In other words, the point cloud corresponding to the large vehicle (vehicle AM3) may not be clustered into a single cluster, but rather split into two or more clusters (cluster C3-1, cluster C3-2, cluster C3-3). The size of each of the split clusters (cluster C3-1, cluster C3-2, cluster C3-3) is smaller than the size of area AA (the rectangular area in the example shown in Figure 8) where the point cloud of the large vehicle (vehicle AM3) is located. If, for each of the split clusters (cluster C3-1, cluster C3-2, cluster C3-3), the object corresponding to that cluster is identified based on the size of the cluster, as detailed below, the object corresponding to that cluster will not be identified as a large vehicle, but rather as an object smaller than a large vehicle (vehicle AM3).

[0033] Therefore, the control unit 120 of the object discrimination device 100 according to this embodiment has a cluster merging processing unit 124. The cluster merging processing unit 124 checks, in each frame of the point cloud information of the detection region A, whether the point cloud corresponding to the same object (for example, a vehicle) is divided into multiple clusters, as detailed below, and if the point cloud corresponding to the same object is divided into multiple clusters, it merges the divided clusters.

[0034] The object discrimination processing unit 125, in each frame of the point cloud information in detection region A, discriminates for each cluster based on the size of the cluster to determine the object corresponding to that cluster.

[0035] The cluster moves in accordance with the movement of the vehicle corresponding to the cluster. Therefore, the direction of movement of the cluster calculated by the direction of movement calculation unit 123 substantially coincides with the direction of movement of the vehicle corresponding to the cluster. As shown in Figure 9, if the width of vehicle AM ​​is WA and its length is LA, and the direction of movement DA of vehicle AM ​​coincides with the direction of movement of the cluster, then, as shown in Figure 10, the length LC of cluster C in the direction of movement corresponds to the length (depth) LA of the vehicle AM ​​corresponding to cluster C, and the length WC in the width direction of cluster C corresponds to the width WA of the vehicle AM ​​corresponding to cluster C. Here, the width direction of cluster C is the direction perpendicular to the direction of movement in the horizontal plane (the plane spanned by the x-axis and y-axis directions) (that is, the direction perpendicular to the direction of movement and the height direction (z-axis direction)).

[0036] Therefore, the object discrimination processing unit 125 determines the object corresponding to the cluster based on, for example, the length LC in the direction of movement of the cluster and the length WC in the width direction of the cluster. At this time, for example, a first numerical range R1 for the length in the width direction of the cluster and a second numerical range R2 for the length in the direction of movement of the cluster are set in advance for each type of object, and the object discrimination processing unit 125 determines the object corresponding to each cluster based on this first numerical range R1, the second numerical range R2, the length WC in the width direction of the cluster and the length LC in the direction of movement of the cluster.

[0037] In other words, the object discrimination processing unit 125 should determine which object corresponds to each cluster based on whether the widthwise length WC of the cluster is included in the first numerical range R1, and whether the length LC of the cluster in the direction of travel is included in the second numerical range R2. For example, if the widthwise length WC of the cluster is included in the first numerical range R1 set for object A, and the length LC of the cluster in the direction of travel is included in the second numerical range R2 set for object A, the object discrimination processing unit 125 will determine that the cluster corresponds to object A.

[0038] The first numerical range R1 and the second numerical range R2 should be set for each type of object based on the size of that type of object. For example, the first numerical range R1 should be set for each type of object based on the width of that type of object, and the second numerical range R2 should be set for each type of object based on the length (depth) of that type of object.

[0039] The first numerical range R1 and the second numerical range R2 are preferably set separately for two-wheeled vehicles (e.g., bicycles and motorcycles) and four-wheeled vehicles (e.g., passenger cars), as shown in Figure 11. In the example shown in Figure 11, the first numerical range R1 for two-wheeled vehicles is set to WTmin ≤ WC ≤ WTmax, and the second numerical range R2 for two-wheeled vehicles is set to LTmin ≤ LC ≤ LTmax. The first numerical range R1 for four-wheeled vehicles is set to WFmin ≤ WC ≤ WFmax, and the second numerical range R2 for four-wheeled vehicles is set to LFmin ≤ LC ≤ LFmax.

[0040] In this case, it is advisable to appropriately set the first numerical range R1 and the second numerical range R2 for motorcycles based on the size of the motorcycle, and to appropriately set the first numerical range R1 and the second numerical range R2 for four-wheeled vehicles based on the size of the four-wheeled vehicle. For example, it is advisable to appropriately set WTmin based on the minimum width of the motorcycle, WTmax based on the maximum width of the motorcycle, LTmin based on the minimum length (depth) of the motorcycle, and LTmax based on the maximum length (depth) of the motorcycle. Also, for example, it is advisable to appropriately set WFmin based on the minimum width of the four-wheeled vehicle, WFmax based on the maximum width of the four-wheeled vehicle, LFmin based on the minimum length (depth) of the four-wheeled vehicle, and LFmax based on the maximum length (depth) of the four-wheeled vehicle.

[0041] Alternatively, two-wheeled vehicles could be divided into bicycles and motorcycles, and a first numerical range R1 and a second numerical range R2 could be set for bicycles based on their size, and a first numerical range R1 and a second numerical range R2 could be set for motorcycles based on their size.

[0042] With respect to four-wheeled vehicles, a first numerical range R1 and a second numerical range R2 may be set for each size. For example, four-wheeled vehicles may be divided into small cars, medium cars, and large cars, and a first numerical range R1 and a second numerical range R2 may be set for small cars based on their size, a first numerical range R1 and a second numerical range R2 may be set for medium cars based on their size, and a first numerical range R1 and a second numerical range R2 may be set for large cars based on their size.

[0043] If, based on the determination of the cluster using the first numerical range R1 and the second numerical range R2 for two-wheeled and four-wheeled vehicles, the cluster is not determined to be either object, the type of the cluster may be determined to be unknown.

[0044] As described above, in this embodiment, in each frame of the point cloud information in detection region A, it is checked whether the point cloud corresponding to the same object (for example, a vehicle) is split into multiple clusters, and if the point cloud corresponding to the same object is split into multiple clusters, the split clusters are combined. Therefore, in this embodiment, object identification is not performed for each split cluster, making it possible to perform object identification more accurately.

[0045] Figure 12 shows an example of a processing operation performed in the control unit 120. The processing operation shown in Figure 12 is performed, for example, each time point cloud information for each frame is acquired by the lidar 110.

[0046] The point cloud information acquisition processing unit 121 acquires point cloud information of the detection area A on the road RW from the lidar 110 (step S1201). The clustering processing unit 122 clusters the point clouds included in the point cloud information acquired by the point cloud information acquisition processing unit 121 into clusters (step S1202). The direction of travel calculation unit 123 calculates the direction of travel for each cluster (step S1203). The cluster merging processing unit 124 checks for each cluster in each frame of the point cloud information of the detection area A whether the point cloud corresponding to the same object has been split into multiple clusters, and if the point cloud corresponding to the same object has been split into multiple clusters, it merges the split clusters (step S1204). The object discrimination processing unit 125 discriminates the object corresponding to each cluster based on the size of the cluster (step S1205).

[0047] <Cluster coupling processing unit 124> The cluster merging processing unit 124 checks whether the point cloud corresponding to the same object is split into multiple clusters, and if it is split into multiple clusters, it merges the split clusters.

[0048] At this time, the cluster merging processing unit 124, in each frame of the point cloud information in detection region A, checks for each cluster, starting with the cluster closest to the lidar 110, whether there is a cluster that may correspond to the same object as the object corresponding to that cluster (for example, a vehicle). If such a cluster exists, it sets the cluster that may correspond to the same object as the object corresponding to that cluster as a group of candidate clusters for merging with that cluster. At this time, for example, the cluster merging processing unit 124 checks for each cluster whether there is a cluster that satisfies the first condition between it and that cluster, and sets the cluster that satisfies the first condition between it and that cluster as a cluster that may correspond to the same object as the object corresponding to that cluster (a cluster in the group of candidate clusters for merging).

[0049] For example, if the detection area A contains only one lane, as shown in Figure 13, the first condition can be set as "the cluster exists within a first length L1 in a first direction DS from the representative point of the cluster." In other words, if the detection area A contains only one lane, the cluster coupling processing unit 124 sets the clusters (cluster C3-2, cluster C3-3) as a group of candidate clusters for coupling with the cluster (cluster C3-1) if multiple clusters (cluster C3-2, cluster C3-3) exist within a first length L1 in a first direction DS from the representative point RP of the cluster (cluster C3-1), as shown in Figure 14. Here, the first direction DS is the direction of travel of the cluster if the point cloud of the cluster spreads out in the direction of travel of the cluster from the representative point of the cluster, and is the opposite direction of travel of the cluster if the point cloud of the cluster spreads out in the opposite direction of travel of the cluster from the representative point of the cluster. The first length L1 is determined, for example, based on the size of a large vehicle. The first length L1 should, for example, be set to the upper limit of the total length of the trailer (18m).

[0050] In the example shown in Figure 14, the point cloud of cluster C3-1 extends from the representative point RP of cluster C3-1 in the opposite direction to the propagation direction DM of cluster C3-1, the first direction DS is the opposite direction to the propagation direction DM of cluster C3-1, and clusters C3-2 and C3-3 are located within a first length L1 in the first direction DS from the representative point RP of cluster C3-1. For this reason, in the example shown in Figure 14, the cluster merging processing unit 124 sets clusters C3-2 and C3-3 as candidate cluster groups for merging with cluster C3-1.

[0051] Furthermore, as shown in Figure 15, if the detection area A includes multiple lanes, the first condition may be defined as "the cluster is located within a first length L1 in a first direction DS from the representative point of the cluster, and within the lane LN to which the object corresponding to the cluster (e.g., a vehicle) is moving." In other words, if the detection area A includes multiple lanes, the cluster coupling processing unit 124, as shown in Figure 16, sets the multiple clusters (cluster C3-2, cluster C3-3) as a group of candidate clusters for coupling with the cluster (cluster C3-1) if they are located within a first length L1 in a first direction DS from the representative point RP of the cluster (cluster C3-1), and within the lane LN to which the object corresponding to the cluster (cluster C3-1) is moving.

[0052] In this case, the cluster merging processing unit 124 may define the lane LN region for each cluster based on the representative point of the cluster, the two points located at the outermost edges in the width direction of the cluster, and a second distance D2. For example, as shown in Figure 17, the cluster coupling processing unit 124 may set the area within a distance D2·(DE1 / (DE1+DE2)) from the representative point RP to the lane LN (area enclosed by two thick dashed lines) on which the object corresponding to the cluster travels, if the distances between the two lines (first endpoint EL1, second endpoint EL2) extending in the direction of travel DM of the cluster, passing through the two points (first endpoint EP1, second endpoint EP2) located at the outermost ends in the width direction of cluster C3-1, and the representative point RP, to be DE1 and DE2, respectively, and the second distance to be D2. In this case, the width of the set lane LN area is the second distance D2 (= D2·(DE1 / (DE1+DE2)) + D2·(DE2 / (DE1+DE2))). Here, the second distance D2 is determined, for example, based on the width of the lane included in detection area A. The second distance D2 may be, for example, the width of a typical lane (3.5m).

[0053] In the example shown in Figure 16, the point cloud of cluster C3-1 extends from the representative point RP of cluster C3-1 in the opposite direction to the direction of travel DM of cluster C3-1, the first direction DS is the opposite direction to the direction of travel DM of cluster C3-1, and clusters C3-2 and C3-3 are located within a first length L1 from the representative point RP of cluster C3-1 in the first direction DS, and within the lane LN (the area enclosed by two thick dashed lines) where the vehicle corresponding to cluster C3-1 travels. For this reason, in the example shown in Figure 16, the cluster merging processing unit 124 sets clusters C3-2 and C3-3 as candidate cluster groups for merging with cluster C3-1.

[0054] The cluster merging processing unit 124 may, for example, determine that if a candidate cluster for merging has been set for each cluster, the candidate cluster group is a group of clusters corresponding to the same object (e.g., a vehicle) as the object corresponding to the cluster, and then merge the clusters included in the candidate cluster group with the cluster.

[0055] In this case, as shown in the examples in Figures 14 and 16, the cluster processing unit 124 determines that the candidate cluster group for joining cluster C3-1 (cluster C3-2, cluster C3-3) is a group of clusters corresponding to the same object (vehicle) as the object corresponding to cluster C3-1, and joins clusters C3-2 and C3-3, which are included in the candidate cluster group, to cluster C3-1.

[0056] As shown in Figure 18, there are cases where a small vehicle (vehicle AM4) is traveling in the same lane as a large vehicle (vehicle AM3) and is located within a first length L1 behind the front of the large vehicle (vehicle AM3). In such cases, as shown in Figure 19, the cluster (cluster C4) corresponding to the small vehicle (vehicle AM4) also exists within a first length L1 in the first direction DR from the representative point RP of the cluster (cluster C3-1) corresponding to the large vehicle (vehicle AM3). Therefore, in such cases, in addition to clusters C3-2 and C3-3 corresponding to the large vehicle (vehicle AM3), cluster C4 is also included in the group of candidate clusters for joining cluster C3-1.

[0057] However, as shown in Figure 18, even if a small vehicle (vehicle AM4) is included within a first length L1 from the front of a large vehicle (vehicle AM3), in an earlier time, as shown in Figure 20, the small vehicle (vehicle AM4) may not be included within a first length L1 from the front of a large vehicle (vehicle AM3). If the small vehicle (vehicle AM4) is not included within a first length L1 from the front of a large vehicle (vehicle AM3), as shown in Figure 20, then, as shown in Figure 21, cluster C4' corresponding to the small vehicle (vehicle AM4) is not included within a first length L1 in the first direction DR' from the representative point RP' of cluster C3-1' corresponding to the large vehicle (vehicle AM3). In that frame, cluster C4' does not satisfy the first condition with cluster C3-1', is not included in the cluster joining candidates of cluster C3-1', and is not joined to cluster C3-1'.

[0058] Therefore, in each frame of the point cloud information in detection region A, the cluster merging processing unit 124 checks, for each cluster, if a group of candidate clusters for merging has been set for that cluster, whether each of the clusters corresponding to the multiple clusters included in the group of candidate clusters for merging in the previous frame (for example, N frames (for example, N=1) prior) is a cluster that may correspond to the same object as the object corresponding to cluster C (for example, a vehicle) (for example, whether the first condition is met in relation to the cluster corresponding to cluster C). Then, the cluster merging processing unit 124 merges the clusters from the group of candidate clusters for merging that correspond to the clusters that may correspond to the same object as the object corresponding to cluster C in the previous frame (i.e., clusters that satisfy the first condition in relation to cluster C' corresponding to cluster C) with cluster C, and does not merge the clusters from the group of candidate clusters that do not correspond to the clusters that may correspond to the same object as the object corresponding to cluster C in the previous frame (i.e., clusters that do not satisfy the first condition in relation to cluster C' corresponding to cluster C) with cluster C.

[0059] In the example shown in Figure 19, the cluster group consisting of clusters C3-2, C3-3, and C4 is set as the candidate cluster group for joining cluster C3-1 in the frame of Figure 19. Then, in the examples shown in Figures 19 and 21, the frame of Figure 21 is a frame prior to the frame of Figure 19, and clusters C3-1', C3-2', C3-3', and C4' in Figure 21 correspond to clusters C3-1, C3-2, C3-3, and C4 in Figure 19, respectively. Then, in the examples shown in Figures 19 and 21, clusters C3-2' and C3-3', which correspond to clusters C3-2 and C3-3 in the frame of Figure 21, exist within a first length L1 in the first direction DR' from the representative point RP' of cluster C3-1', and satisfy the first condition, but cluster C4', which corresponds to cluster C4, does not exist within a first length L1 in the first direction DR', and does not satisfy the first condition. Therefore, in the example shown in Figures 19 and 21, the cluster merging processing unit 124 merges clusters C3-2 and 3-3 from the group of candidate clusters to merge into cluster C3-1, but does not merge cluster 4 into cluster C3-1.

[0060] Figure 22 shows an example of the processing operation performed by the cluster coupling processing unit 124 in step S1204 of Figure 12. As described above, the processing operation shown in Figure 12 is performed, for example, each time point cloud information for each frame is acquired by the lidar 110. The processing operation shown in Figure 22 is performed in step S1203 of Figure 12 for each cluster that was clustered in step S1201. At this time, the processing operation shown in Figure 22 is performed, for example, starting with the clusters whose representative point is closest to the lidar 110.

[0061] The system checks whether there is a cluster that satisfies the first condition with cluster C (step S2201). If no such cluster exists (step S2201, NO), the process terminates. If such a cluster exists (step S2201, YES), the cluster that satisfies the first condition with cluster C is set as a group of candidate clusters for joining to cluster C (step S2202).

[0062] In the frame preceding the current frame, the cluster merging unit 124 checks whether each of the clusters corresponding to multiple clusters included in the group of candidate clusters satisfies the first condition with respect to cluster C' corresponding to cluster C in the frame preceding the current frame. The cluster merging unit 124 then merges the clusters from the group of candidate clusters that satisfy the first condition with respect to cluster C' corresponding to cluster C in the frame preceding the current frame into cluster C, and does not merge the clusters from the group of candidate clusters that do not satisfy the first condition with respect to cluster C' corresponding to cluster C in the frame preceding the current frame into cluster C (step S2203).

[0063] <Using cluster height> Large and small vehicles have different heights. For example, a vehicle with a length of 4.7m or less is a small vehicle and is unlikely to be a large vehicle. Therefore, the first length L1 used when setting the candidate cluster group for joining may be changed based on the height length HC of the cluster. For example, if the point cloud information of the detection area A acquired by the point cloud information acquisition processing unit 121 is three-dimensional point cloud information, the cluster joining processing unit 124 checks whether the height length HC of each cluster is greater than the first height H1 (for example, the upper limit of the height of a small vehicle (for example, 2m)). If the height length HC of the cluster is greater than or equal to the first height H1, the first length L1 is set to the first value V1. If the height length HC of the cluster is not greater than or equal to the first height H1, the first length L1 is set to the second value V2. Here, the first value V1 is determined based on the size of a large vehicle, for example. The first value V1 may be set to the upper limit of the total length of a trailer (18m), for example. Furthermore, the second value V2 is smaller than the first value V1 and is determined, for example, based on the size of a small vehicle. The second value V2 should ideally be set to, for example, the upper limit of the vehicle length for a small vehicle (4.7m).

[0064] Figure 23 shows an example of the processing operation performed by the cluster coupling processing unit 124 in step S1204 of Figure 12. As described above, the processing operation shown in Figure 12 is performed, for example, each time point cloud information for each frame is acquired by the lidar 110. The processing operation shown in Figure 23 is performed in step S1203 of Figure 12 for each cluster that was clustered in step S1201. At this time, the processing operation shown in Figure 23 is performed, for example, starting with the clusters whose representative point is closest to the lidar 110.

[0065] The system checks whether the height length HC of cluster C is greater than the first height H1 (step S2301). If the height length HC of cluster C is greater than or equal to the first height H1 (step S2301, YES), the first length L1 is set to the first value V1 (step S2302). If the height length HC of cluster C is not greater than or equal to the first height H1 (step S2301, NO), the first length L1 is set to the second value V2 (step S2303).

[0066] The system checks whether there is a cluster that satisfies the first condition with cluster C (step S2304). If no such cluster exists (step S2304, NO), the process terminates. If such a cluster exists (step S2304, YES), the cluster that satisfies the first condition with cluster C is set as a group of candidate clusters for joining to cluster C (step S2305).

[0067] In the frame preceding the current frame, the cluster joining processing unit 124 checks whether each of the clusters corresponding to multiple clusters included in the group of candidate clusters satisfies the first condition with respect to cluster C', which corresponds to cluster C, in the frame preceding the current frame. The cluster joining processing unit 124 then joins cluster C to cluster C if it corresponds to a cluster in the group of candidate clusters that satisfies the first condition with respect to cluster C', which corresponds to cluster C, in the frame preceding the current frame, and does not join cluster C' to cluster C' (step S2306).

[0068] Furthermore, the cluster merging processing unit 124 may, if the height length HC of a cluster is greater than or equal to the first height H1, check whether the point cloud corresponding to the same vehicle is split into multiple clusters, as described above, and if the height length HC of a cluster is less than the first height H1, it may not check whether the point cloud corresponding to the same vehicle is split into multiple clusters.

[0069] Figure 24 shows an example of a processing operation performed by the cluster coupling processing unit 124 in step S1204 of Figure 12. As described above, the processing operation shown in Figure 12 is performed, for example, each time point cloud information for each frame is acquired by the lidar 110. The processing operation shown in Figure 24 is performed in step S1203 of Figure 12 for each cluster that was clustered in step S1201. At this time, the processing operation shown in Figure 24 is performed, for example, starting with the clusters whose representative point is closest to the lidar 110.

[0070] The process checks whether the height length HC of cluster C is greater than the first height H1 (step S2401). If the height length HC of cluster C is not greater than or equal to the first height H1 (step S2401, NO), the process is terminated. If the height length HC of cluster C is greater than or equal to the first height H1 (step S2401, YES), the process checks whether there is a cluster between cluster C and the other cluster that satisfies the first condition (step S2402).

[0071] If no cluster satisfies the first condition with cluster C (step S2402, NO), the process is terminated. If such a cluster exists (step S2402, YES), the clusters that satisfy the first condition with cluster C are set as a group of candidate clusters for joining to cluster C (step S2403).

[0072] In the frame preceding the current frame, the cluster joining processing unit 124 checks whether each of the clusters corresponding to multiple clusters included in the group of candidate clusters satisfies the first condition with respect to cluster C', which corresponds to cluster C, in the frame preceding the current frame. The cluster joining processing unit 124 then joins cluster C to cluster C if it corresponds to a cluster in the group of candidate clusters that satisfies the first condition with respect to cluster C', which corresponds to cluster C, in the frame preceding the current frame, and does not join cluster C to cluster C (step S2404).

[0073] The present invention has been described above with reference to preferred embodiments. While the present invention has been described with specific examples, various modifications and changes can be made to these examples without departing from the spirit and scope of the invention as described in the claims. [Explanation of Symbols]

[0074] 100 Object discrimination device 110 Rider 120 Control Unit 121 Point Cloud Information Acquisition Processing Unit 122 Clustering Processing Unit 123 Traveling direction calculation unit 124 Cluster Binding Processing Unit 125 Object Recognition Processing Unit

Claims

1. A point cloud information acquisition processing unit that acquires point cloud information of the detection area on the road, A clustering processing unit that clusters the point clouds included in the point cloud information into clusters, An information processing device comprising: a cluster merging processing unit that checks whether a point cloud corresponding to the same object is split into multiple clusters, and if so, merges the split clusters.

2. The information processing apparatus according to claim 1, wherein the cluster joining processing unit joins each of the clusters with a cluster that satisfies a first condition with respect to that cluster.

3. The information processing apparatus according to claim 2, wherein in each of the clusters, the first condition is that it is located within a first length in a first direction from a representative point of the cluster.

4. The information processing apparatus according to claim 2, wherein in each of the clusters, the first condition is located within a first length in a first direction from a representative point of the cluster and within the lane in which the object corresponding to the cluster moves.

5. The cluster coupling processing unit sets the lane region for each of the clusters based on the representative point of the cluster and the two points located at the outermost edges in the width direction of the cluster. The information processing apparatus according to claim 4, wherein the width direction of the cluster is perpendicular to the direction of movement of the cluster in the horizontal plane.

6. The information processing apparatus according to any one of claims 3 to 5, wherein in each of the clusters, the first direction is the direction of progression of the cluster if the point cloud of the cluster spreads out from the representative point of the cluster in the direction of progression of the cluster, and is the opposite direction of progression of the cluster if the point cloud of the cluster spreads out from the representative point of the cluster in the opposite direction of progression of the cluster.

7. The information processing apparatus according to any one of claims 3 to 5, wherein the first length is determined based on the size of a large vehicle.

8. The point cloud information consists of multiple frames generated at a predetermined frame rate. The clustering processing unit clusters the point cloud into clusters in each frame of the point cloud information. The cluster joining processing unit, in each frame of the point cloud information, performs the following for each cluster that has been clustered in that frame: Clusters that satisfy the first condition with the cluster in question are set as candidate clusters for joining. If the group of candidate clusters for joining is set for the cluster, the information processing apparatus according to any one of claims 2 to 5, which joins the cluster corresponding to the cluster that satisfies the first condition in the frame preceding the frame corresponding to the cluster, from among the group of candidate clusters for joining, to the cluster.

9. The information processing apparatus according to any one of claims 3 to 5, wherein the cluster coupling processing unit sets the first length to a first value if the height of each cluster is equal to or greater than a first height, and sets the first length to a second value which is smaller than the first value if the height of the cluster is not equal to or greater than a first height.

10. The information processing apparatus according to any one of claims 2 to 5, wherein the cluster joining processing unit joins each of the clusters with a cluster that satisfies the first condition with the cluster, if the height of the cluster is equal to or greater than a first height.

11. The information processing apparatus according to any one of claims 1 to 5, further comprising an object discrimination processing unit for each of the aforementioned clusters, which determines the object corresponding to the cluster based on the size of the cluster.

12. A method of information processing performed by a computer, A point cloud information acquisition process that acquires point cloud information of the detection area on the road, A clustering process is performed to cluster the point clouds included in the aforementioned point cloud information into clusters. An information processing method comprising: checking whether a point cloud corresponding to the same object is divided into multiple clusters; and, if the point cloud corresponding to the same object is divided into multiple clusters, a cluster merging process that merges the divided clusters.

13. An information processing program that causes a computer to execute the information processing method described in claim 12.

Images