Object recognition method and object recognition device

Abstract

To provide an object recognition method and an object recognition apparatus configured to improve the accuracy of estimating attitude of a target object.SOLUTION: An object recognition method includes: extracting, as a ranging point group, multiple pieces of ranging data on a surface of the same target object, out of pieces of ranging point data on the surface of the target object existing around a mobile body, the ranging point data being generated by a ranging sensor 10 mounted on the mobile body; converting the ranging point group onto a two-dimensional coordinate or a three-dimensional coordinate having an origin at a position of the ranging sensor; setting multiple model areas including all of the ranging point group on the two-dimensional coordinate and the three-dimensional coordinate; setting, as an occupied area, an area which is a blind area for the ranging sensor due to the ranging point group; setting, as a non-occupied area, an area other than the occupied area; selecting a model area, out of the multiple model areas, in which the area or the volume of an overlap area where each of the model areas overlaps the non-occupied area is equal to or smaller than a predetermined value; and recognizing the selected model area as a target object.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to an object recognition method and an object recognition device. [Background technology]

[0002] Patent Document 1 discloses an object recognition device that identifies the type of a target object by estimating the rectangle of the target object from information on the detection pattern of a group of reflection points of the target object detected by an on-board radar. In this rectangle estimation, all reflection points of the detection pattern obtained by clustering are rotated one degree at a time on a horizontal plane on which each reflection point is plotted, with the center of the cluster as the origin, and the density of the reflection points in the horizontal direction is projected each time. Then, from the rotation state where the density value of the reflection point peaks, an axis that forms at least one of the two L-shaped sides of the rectangle is found. If only one axis is found, the other axis that forms the remaining one of the two L-shaped sides is set in a direction perpendicular to the axis that forms the original axis, thereby detecting the axis that forms the width or depth of the rectangle. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2012-52838 Summary of the Invention [Problem to be solved by the invention]

[0004] However, if there is a reflection point inside the contour of the target object on a horizontal plane (top view), the peak of the density value may deviate from the side that corresponds to the actual width or depth of the target object. In particular, if the target object is tilted relative to the horizontal direction, the reflection points on two sides of the L-shape of the target object may overlap, causing the peak of the density value to deviate from the side that corresponds to the actual width or depth of the target object. This may result in erroneous detection of the axis that corresponds to the width or depth of the rectangle, reducing the accuracy of estimating the target object's posture.

[0005] An object of the present invention is to improve the accuracy of estimating the posture of a target object. [Means for solving the problem]

[0006] An object recognition method and object recognition device according to one aspect of the present invention comprises a ranging sensor mounted on a moving body that generates ranging point data relating to ranging points on the surface of a target object present around the moving body, and a controller that recognizes the target object based on the ranging point data acquired from the ranging sensor. The controller extracts multiple ranging point data on the surface of the same target object from the ranging point data as a ranging point cloud, converts the ranging point cloud onto two-dimensional or three-dimensional coordinates with the position of the ranging sensor as the origin, sets multiple model areas including all of the ranging point clouds on the two-dimensional or three-dimensional coordinates, sets areas on the two-dimensional or three-dimensional coordinates that are blind spots from the ranging sensor due to the ranging point cloud as occupied areas, sets areas on the two-dimensional or three-dimensional coordinates other than the occupied areas as non-occupied areas, selects from the multiple model areas a model area where the area or volume of an overlapping area where each of the model areas overlaps with the non-occupied area is less than a predetermined value, and recognizes the selected model area as the target object. [Effects of the Invention]

[0007] According to the present invention, the accuracy of estimating the posture of a target object is improved. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a configuration diagram of an object recognition device according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram for explaining an example of a method for converting distance measurement point data on the surface of another vehicle into two-dimensional distance measurement point cloud data viewed from above the moving body. [Figure 3] FIG. 3 is a diagram showing distance measurement point data between a nearby vehicle and a wall, and between the nearby vehicle and the surface of the wall. [Figure 4]FIG. 4 is a diagram illustrating an example of a method for setting a model region. [Figure 5] FIG. 5 is a diagram for explaining a method for setting an unoccupied area. [Figure 6] FIG. 6 is a diagram illustrating a method for selecting a model region. [Figure 7] FIG. 7 is a flowchart showing the operation of the object recognition device according to the first embodiment. [Figure 8] FIG. 8 is a flowchart showing the operation of the object recognition device according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] (First embodiment) A first embodiment of the present invention will be described below with reference to the drawings. In the description, the same components are designated by the same reference numerals and duplicated explanations will be omitted.

[0010] An example of the configuration of an object recognition device 1 according to the first embodiment will be described with reference to Fig. 1. As shown in Fig. 1, the object recognition device 1 includes a distance measurement sensor 10, and a control unit 20 including a distance measurement point acquisition unit 21, a clustering processing unit 22, a surrounding object recognition unit 23, and a driving assistance unit 24.

[0011] The ranging sensor 10 is mounted on a mobile body and generates ranging point data relating to ranging points on the surface of target objects present around the mobile body. The ranging sensor 10 transmits probe waves around the mobile body to scan the periphery of the mobile body and receives reflected waves of the probe waves reflected by the surfaces of target objects present around the mobile body. Based on the received reflected waves, the ranging sensor 10 calculates the positions of the reflection points where the probe waves are reflected at multiple positions on the surface of the target object as the relative positions of the reflection points with respect to the ranging sensor 10 of the mobile body, and obtains ranging point data indicating the relative positions of each reflection point.

[0012] An example of the distance measurement sensor 10 is a LIDAR (Laser Imaging Detection and Ranging). A LIDAR measures the distance and direction to an object and recognizes the shape of the object by emitting light (laser light) to a surrounding object and measuring the time it takes for the light (reflected light) to hit the object and bounce back. Furthermore, a LIDAR can obtain the positional relationship of objects in three dimensions. It is also possible to perform mapping using the intensity of the reflected light.

[0013] The ranging point data includes position information of the ranging point. The position information of the ranging point is information indicating the position coordinates of the ranging point. For the position coordinates, a three-dimensional coordinate system expressed by x-coordinates, y-coordinates, and z-coordinates with the installation position of the ranging sensor 10 as the origin is used. For the position coordinates, a polar coordinate system expressed by the direction (yaw angle, pitch angle) from the ranging sensor 10 to the ranging point and the distance (depth) from the ranging sensor 10 to the ranging point may be used. Furthermore, the ranging point data is acquired at predetermined intervals.

[0014] The distance measurement sensor 10 may be any sensor capable of acquiring the positions of multiple distance measurement points on the surface of a target object as distance measurement point data, and is not limited to the above. For example, the distance measurement sensor 10 may be a stereo camera mounted on a moving object. When a stereo camera is used as the distance measurement sensor 10, at least two camera images from different viewpoints are acquired, parallax is calculated from corresponding positions between the images, and three-dimensional information can be obtained for each position in the image using the principle of triangulation. The distance measurement point data may also be acquired from the positions and distances of pixels corresponding to surrounding objects in images captured by the stereo camera around the moving object. The distance measurement sensor 10 outputs the acquired distance measurement point data to the distance measurement point acquisition unit 21 of the control unit 20.

[0015] The control unit 20 is a general-purpose microcomputer equipped with a CPU (Central Processing Unit), memory, and input / output units. A computer program for causing the microcomputer to function as the object recognition device 1 is installed in the microcomputer. By executing the computer program, the microcomputer functions as multiple information processing circuits equipped in the object recognition device 1. The control unit 20 processes data acquired from the distance measurement sensor 10.

[0016] Although an example is shown here in which the multiple information processing circuits provided in the object recognition device 1 are realized by software, it is of course also possible to configure the information processing circuits by providing dedicated hardware for executing each of the information processes described below.Furthermore, the multiple information processing circuits may be configured by individual hardware.

[0017] The control unit 20 recognizes a target object based on the ranging point data acquired by the ranging sensor 10. The control unit 20 includes a ranging point acquisition unit 21, a clustering processing unit 22, a surrounding object recognition unit 23, and a driving assistance unit 24 as examples of a plurality of information processing circuits (information processing functions).

[0018] Each information processing function of the control unit 20 will be described below. The ranging point acquisition unit 21 acquires ranging point data from the ranging sensor 10. The ranging point data acquired by the ranging sensor 10 includes various ranging point data of road surfaces and the like other than objects. Therefore, the ranging point acquisition unit 21 first extracts ranging point data of three-dimensional objects from the acquired ranging point data.

[0019] The method for extracting distance measurement point data of a three-dimensional object may extract distance measurement point data at a height equal to or greater than a first predetermined height value from the road surface. For example, the first predetermined height value may be set to the height of the bottom of a typical vehicle bumper on the z-axis of a three-dimensional coordinate system. Alternatively, the method for extracting distance measurement point data of a three-dimensional object may extract distance measurement point data at a height equal to or greater than a first predetermined height value and less than a second predetermined height value from the road surface. For example, the second predetermined height value may be set to include the vehicle type (large vehicle, standard vehicle, etc.) of the parked vehicle recognized by a camera or the like, and the roof height corresponding to the vehicle type. Alternatively, a grid map storing distance measurement point data may be generated to determine the distance measurement points between the road surface and the three-dimensional object.

[0020] Next, the clustering processing unit 22 performs a clustering process to extract ranging point data of the extracted three-dimensional object. Here, clustering is a technique for extracting, from the ranging point data of the extracted three-dimensional object, multiple ranging point data on the surface of the same target object as a cluster (ranging point cloud). As an example, from the multiple ranging point data of the extracted three-dimensional object, ranging point data in which the distance between two points is equal to or less than a predetermined value is extracted as a ranging point cloud. For example, when the target object is a vehicle (another vehicle), the extracted ranging point cloud will look like (a) in FIG. 2.

[0021] FIG. 3 shows distance measurement points between a nearby vehicle and a wall, and between the vehicle and the surface of the wall. When the vehicle is close to a wall, distance measurement points of multiple target objects, such as the vehicle and the wall, may be erroneously extracted as distance measurement points of the same target object. In such cases, the clustering processing unit 22 reduces the predetermined value of the distance between the two points used to extract the distance measurement point cloud from the previous value, performs clustering processing again, and re-extracts the distance measurement point cloud. In other words, the clustering processing unit 22 divides the distance measurement point cloud. A specific flow will be described later using the flowchart in FIG. 7.

[0022] Next, the clustering processing unit 22 converts the extracted ranging point cloud into two-dimensional ranging point cloud data viewed from above of the moving object, as shown in Fig. 2(b). Here, the two-dimensional ranging point cloud data viewed from above of the moving object refers to ranging point cloud data obtained by projecting the ranging point cloud onto a two-dimensional coordinate system expressed by x-coordinates and y-coordinates with the position of the ranging sensor 10 as the origin and converting it into two-dimensional coordinates, and this meaning will be used in the following. Note that the clustering processing unit 22 may also project the ranging point cloud onto a three-dimensional coordinate system expressed by x-coordinates, y-coordinates, and z-coordinates with the position of the ranging sensor as the origin and convert it into three-dimensional coordinates.

[0023] Next, the surrounding object recognition unit 23 sets a model region of the target object based on the two-dimensional ranging point cloud data input from the clustering processing unit 22. The surrounding object recognition unit 23 sets a plurality of model regions that include all of the ranging point clouds on the two-dimensional coordinate system. Since most objects (cars, motorcycles, bicycles, etc.) that exist in a traffic environment have a shape that is close to a rectangle when viewed from above, the surrounding object recognition unit 23 sets a rectangular region that indicates a rectangular shape that approximates the target object as a model region based on the assumption that the target object has a rectangular shape.

[0024] When the ranging point cloud is converted into three-dimensional coordinates, the surrounding object recognition unit 23 may set a rectangular parallelepiped region that approximates the target object as a model region on the three-dimensional coordinates, based on the assumption that the target object has a rectangular parallelepiped shape. In other words, the surrounding object recognition unit 23 sets multiple model regions that include all of the ranging point cloud on the two-dimensional coordinates or three-dimensional coordinates.

[0025] As an example of a method for setting a model area, the surrounding object recognition unit 23 may set rectangular areas (model areas) 30a, 30b, and 30c of minimum area that include all of the ranging point cloud, with different angles θ between the long side direction of the rectangle and the traveling direction of the target object V2, as shown in (a), (b), and (c) of FIG. 4. In (b) of FIG. 4, the angle θ between the long side direction of the rectangle and the traveling direction of the target object V2 is 0°. For example, the surrounding object recognition unit 23 may set multiple model areas of minimum area that include all of the ranging point cloud, with the angle θ between the long side direction of the rectangle and the traveling direction of the target object varying in 1° increments from 0° to 89°. In this case, 90 different rectangular areas with angles θ between the long side direction of the rectangle and the traveling direction of the target object varying from 0° to 89° are set as model areas. The number of degrees at which the model areas are set is not limited to the above. In this embodiment, an example has been shown in which a model region with the smallest area that includes all of the range-finding points is set, but a model region that includes some of the range-finding points may also be set.

[0026] When the ranging point cloud is converted into three-dimensional coordinates, the surrounding object recognition unit 23 may set multiple model areas of minimum volume that include all of the ranging point cloud on the three-dimensional coordinates, where the angle between the long side direction of the rectangular parallelepiped on the xy plane and the traveling direction of the target object is different. In this case, 90 different rectangular parallelepiped areas, where the angle between the long side direction of the rectangular parallelepiped on the xy plane and the traveling direction of the target object ranges from 0° to 89°, may be set as model areas.

[0027] 5, it is assumed that the distance measurement sensor 10 is mounted on a moving object V1 and that a distance measurement point cloud of a target object V2 has been extracted. The surrounding object recognition unit 23 sets an area that is a blind spot from the distance measurement sensor 10 (origin) based on the distance measurement point cloud as an occupied area 40 on a two-dimensional coordinate system with the position of the distance measurement sensor 10 as the origin, as shown in FIG. 5. The surrounding object recognition unit 23 also sets an area other than the occupied area 40 as a non-occupied area 50 on the two-dimensional coordinate system with the position of the distance measurement sensor 10 as the origin. Note that when the distance measurement point cloud is converted into three-dimensional coordinates, the surrounding object recognition unit 23 may set an area that is a blind spot from the distance measurement sensor 10 (origin) based on the distance measurement point cloud as an occupied area on the three-dimensional coordinate system with the position of the distance measurement sensor 10 as the origin, and set an area other than the occupied area as a non-occupied area on the three-dimensional coordinate system.

[0028] Here, the occupied area is the area where the probe wave is blocked by the ranging point group when the probe wave is applied, with the position of the ranging sensor 10 as the origin. Note that the ranging point groups are considered to have continuous points. The occupied area is set based on the ranging point of the ranging point group that is the shortest distance from the ranging sensor 10 in each direction from the ranging sensor 10.

[0029] The surrounding object recognition unit 23 calculates the area of ​​the overlapping region where each of the multiple model regions overlaps with the non-occupied region 50. For example, the area of ​​the overlapping region where the model region 30a in FIG. 4(a) overlaps with the non-occupied region 50 is the area of ​​the shaded region Ra in FIG. 6(a). The area of ​​the overlapping region where the model region 30b in FIG. 4(b) overlaps with the non-occupied region 50 does not exist as shown in FIG. 6(b), so the area of ​​the overlapping region is 0. The area of ​​the overlapping region where the model region 30c in FIG. 4(c) overlaps with the non-occupied region 50 is the area of ​​the shaded region Rc in FIG. 6(c). Note that when the ranging point cloud is converted into three-dimensional coordinates, the surrounding object recognition unit 23 calculates the volume of the overlapping region where each of the multiple model regions overlaps with the non-occupied region.

[0030] The surrounding object recognition unit 23 selects a model area where the area or volume of the overlapping area is equal to or less than a predetermined value. When there are multiple model areas where the area or volume of the overlapping area is equal to or less than a predetermined value, the surrounding object recognition unit 23 may select the model area where the area or volume of the overlapping area is the smallest. For example, in FIG. 6, the surrounding object recognition unit 23 may select the model area 30b (not shown) where the area of ​​the overlapping area is the smallest (0). Note that the method of selecting a model area is not limited to the above method, and the surrounding object recognition unit 23 may select a model area where the area or volume of the overlapping area where each of the multiple model areas overlaps with the occupied area is equal to or greater than a predetermined value.

[0031] The surrounding object recognition unit 23 recognizes the selected model area as a target object existing around the moving body. In other words, it recognizes the position, shape, and posture of the selected model area as the position, shape, and posture of the target object. If the width of the model area is close to the width of the vehicle, that width may be estimated as the front or rear of the vehicle. Furthermore, by using map information, it is possible to estimate whether the vehicle is facing forward or backward from the lane in which the vehicle is located. The attributes of the recognized target object can be determined based on at least one of the size and shape of the model area. For example, the vehicle type (whether it is a truck or a passenger car) can be determined based on the size of the model area.

[0032] The driving assistance unit 24 provides optimal driving assistance to the moving body equipped with the distance measurement sensor 10 based on information about the target object recognized by the surrounding object recognition unit 23. For example, when the driving assistance unit 24 determines that another vehicle is approaching the vehicle when the distance measurement sensor 10 is installed in the vehicle, the driving assistance unit 24 controls various actuators of the vehicle, such as the steering actuator, accelerator pedal actuator, and brake actuator, to stop or decelerate the vehicle or to provide driving assistance involving avoidance steering. In this case, the accuracy of estimating the vehicle type of the target object improves, thereby improving the accuracy of driving assistance.

[0033] Next, an example of the operation of the object recognition device 1 according to the first embodiment will be described with reference to the flowchart of FIG.

[0034] 7, the distance measurement sensor 10 generates distance measurement point data relating to distance measurement points on the surface of a target object present around the moving body. The distance measurement sensor 10 outputs the generated distance measurement point data to the distance measurement point acquisition unit 21. The distance measurement point acquisition unit 21 acquires the distance measurement point data from the distance measurement sensor 10.

[0035] In step S102, the ranging point acquisition unit 21 extracts ranging point data of the three-dimensional object from the ranging point data acquired in step S101.

[0036] In step S103, the clustering processing unit 22 performs clustering processing on the ranging point data of the three-dimensional object extracted in step S102, and extracts ranging point clouds of the same target object. The clustering processing unit 22 extracts, from the ranging point data of the three-dimensional object, multiple ranging point data on the surface of the same target object as a ranging point cloud.

[0037] In step S104, the clustering processing unit 22 converts the ranging point cloud extracted in step S103 into two-dimensional coordinates with the position of the ranging sensor 10 as the origin. That is, the ranging point cloud extracted in step S103 is converted into two-dimensional ranging point cloud data.

[0038] In step S105, the surrounding object recognition unit 23 sets a plurality of model regions based on the two-dimensional ranging point cloud data obtained in step S104. The surrounding object recognition unit 23 sets a plurality of model regions of the smallest area that include all of the ranging point cloud on the two-dimensional coordinate system. For example, on the two-dimensional coordinate system, the surrounding object recognition unit 23 sets rectangular regions of the smallest area that include all of the ranging point cloud for each 1° angle between the long side direction of the rectangle and the traveling direction of the target object from 0° to 89° as model regions.

[0039] In step S106, the surrounding object recognition unit 23 sets an occupation area based on the two-dimensional ranging point cloud data acquired in step S104. On a two-dimensional coordinate system with the position of the ranging sensor 10 as the origin, the surrounding object recognition unit 23 sets an area that is a blind spot from the ranging sensor 10 (origin) based on the ranging point cloud as the occupation area.

[0040] In step S107, the surrounding object recognition unit 23 sets a non-occupied area based on the two-dimensional ranging point cloud data acquired in step S104. The surrounding object recognition unit 23 sets an area other than the occupied area as the non-occupied area on a two-dimensional coordinate system with the position of the ranging sensor 10 as the origin.

[0041] In step S108, the surrounding object recognition unit 23 calculates the area of ​​the overlap region where each of the plurality of model regions obtained in step S105 overlaps with the unoccupied region obtained in step S107.

[0042] In step S109, the surrounding object recognition unit 23 determines whether or not there is a model area in which the area of ​​the overlapping area is equal to or less than a predetermined value. If there is a model area in which the area of ​​the overlapping area is equal to or less than the predetermined value, the process proceeds to step S111; if there is no model area in which the area of ​​the overlapping area is equal to or less than the predetermined value, the process proceeds to step S110.

[0043] In step S110, the clustering processing unit 22 divides the ranging point cloud. Specifically, the clustering processing unit 22 reduces the predetermined value of the distance between the two points from which the ranging point cloud was extracted from compared to the previous value. Then, the flow returns to step S104. That is, if there is no model region where the area or volume of the overlapping region is equal to or less than the predetermined value, the clustering processing unit 22 divides the ranging point cloud and re-extracts ranging points on the surface of the same target object from the divided ranging point cloud as a ranging point cloud.

[0044] In step S111, the surrounding object recognition unit 23 determines whether there are multiple model regions whose overlapping area is equal to or less than a predetermined value. If there are multiple model regions whose overlapping area is equal to or less than the predetermined value, the process proceeds to step S113. If there are not multiple model regions whose overlapping area is equal to or less than the predetermined value, the process proceeds to step S112. In step S112, the surrounding object recognition unit 23 selects a rectangular region whose overlapping area is equal to or less than the predetermined value.

[0045] In step S113, the surrounding object recognition unit 23 selects a model region in which the area of ​​the overlapping region is smallest.

[0046] In step S114, the surrounding object recognition unit 23 recognizes the model area selected in step S112 or step S113 as the target object. The control unit 20 ends the processing of Fig. 7, and the driving assistance unit 24 provides optimal driving assistance to the mobile object on which the distance measuring sensor 10 is mounted, based on the information on the target object recognized in step S114.

[0047] (Effects of the first embodiment) (1) In the first embodiment, a target object is recognized based on ranging point data related to ranging points on the surface of a target object existing around the mobile object, generated by a ranging sensor 10 mounted on the mobile object. Among the ranging point data, multiple ranging point data on the surface of the same target object are extracted as a ranging point cloud. The ranging point cloud is converted into a two-dimensional or three-dimensional coordinate system with the position of the ranging sensor as the origin. Multiple model areas including all of the ranging point clouds are set on the two-dimensional or three-dimensional coordinate system, and areas blind spots from the ranging sensor 10 due to the ranging point clouds are set as occupied areas, and areas other than the occupied areas are set as unoccupied areas. From the multiple model areas, model areas are selected in which the area or volume of the overlapping area where each model area and the unoccupied area overlap is equal to or smaller than a predetermined value, and the selected model area is recognized as the target object.

[0048] Of the multiple model regions that have been set, model regions whose overlapping areas with unoccupied regions have an area or volume that is less than a predetermined value can be selected and recognized as objects surrounding the target. This allows model regions whose deviation from the target object is less than a predetermined value to be selected from the multiple model regions that have been set from the ranging point data. By recognizing the selected model region as the target object, the accuracy of estimating the posture of the target object can be improved.

[0049] (2) In the first embodiment, on a two-dimensional coordinate system or a three-dimensional coordinate system, an occupied area is set based on the ranging point among the ranging points that has the shortest distance from the ranging sensor 10 in each direction from the ranging sensor 10. The ranging point at which the distance from the ranging sensor is shortest in each direction from the ranging sensor 10 is the ranging point on the contour of the target object, so the occupied area can be set based on the ranging point on the contour of the target object. This allows the occupied area to be set accurately without being affected by ranging points inside the contour of the target object, and the unoccupied area to be set accurately, thereby improving the accuracy of estimating the orientation of the target object.

[0050] (3) In the first embodiment, if there are multiple model regions whose overlapping areas have an area or volume equal to or less than a predetermined value, the model region with the smallest overlapping area or volume is selected and recognized as the target object. This allows the optimum model region to be selected, and the posture of the target object can be estimated based on the optimum model region.

[0051] (4) In the first embodiment, if there is no model area where the area of ​​the overlapping area is less than a predetermined value, the ranging point group is divided, and from the divided ranging point group, ranging points on the surface of the same target object are re-extracted as the ranging point group. This allows the clustering of distance measurement points for a plurality of target objects to be redone even if the distance measurement points are mistakenly clustered as a single target object distance measurement point cloud.

[0052] (Second embodiment) A second embodiment of the present invention will be described below. The second embodiment aims to stably maintain the estimation accuracy of the posture of a target object even when the estimation accuracy of the model region decreases. Note that elements similar to those in the first embodiment are given the same reference numerals, and their description will be omitted. Differences between the second embodiment of the present invention and the first embodiment will be described below.

[0053] In the object recognition device 1 according to the first embodiment, when there are multiple model regions whose overlapping area or volume is equal to or less than a predetermined value, the surrounding object recognition unit 23 selects the model region whose overlapping area is the smallest in area or volume. However, in the object recognition device 1 according to the second embodiment, when there are multiple model regions whose overlapping area or volume is equal to or less than a predetermined value, the surrounding object recognition unit 23 selects a model region whose angular difference from the model region selected when the target object was recognized one frame earlier is equal to or less than a predetermined angular difference, and recognizes the model region as the target object. In other words, when there are multiple model regions whose overlapping area or volume is equal to or less than a predetermined value, the surrounding object recognition unit 23 selects a rectangular region whose angular difference from the model region selected during the previous processing and recognized as the target object is equal to or less than a predetermined angular difference.

[0054] FIG. 8 is a flowchart illustrating an example of the operation of the object recognition device 1 according to the second embodiment. As shown in the figure, in this embodiment, the processes of steps S201 to S212 and step S214 in FIG. 8 are the same as the processes of steps S101 to S112 and step S114 in FIG. 7, and therefore description thereof will be omitted. The process of step S213 in FIG. 8 differs from the first embodiment. The object recognition device 1 according to this embodiment repeatedly executes a series of processes consisting of steps S201 to S214 shown in FIG. 8 at a predetermined cycle. The unit of repeatedly executed processes is called a "frame." In other words, the object recognition device 1 according to the second embodiment repeatedly executes the frames shown in FIG. 8, and selects a model region to be recognized as a target object for each frame.

[0055] In step S213, the surrounding object recognition unit 23 selects a model area whose angular difference from the model area selected in the previous processing and recognized as the target object is equal to or less than a predetermined angular difference.

[0056] (Effects of the second embodiment) (5) In the second embodiment, if there are multiple model regions whose overlapping areas have an area or volume that is less than a predetermined value, a model region whose angular difference with the model region selected when the target object was recognized one frame earlier is less than a predetermined angular difference is selected and recognized as the target object. This makes it possible to stably maintain the accuracy of estimating the posture of the target object even if the estimation accuracy of the model region drops momentarily. [Explanation of symbols]

[0057] 10. Distance sensor 20 Control unit (controller) 21 Focus point acquisition section 22 Clustering processing section 23 Surrounding object recognition unit 24 Driving Support Department

Claims

1. a distance measurement sensor mounted on a moving body and generating distance measurement point data relating to distance measurement points on the surface of a target object present around the moving body; a controller that recognizes the target object based on the distance measurement point data acquired from the distance measurement sensor, The controller extracting a plurality of pieces of distance measurement point data on the surface of the same target object from the distance measurement point data as a distance measurement point group; converting the range-finding point cloud into two-dimensional or three-dimensional coordinates with the position of the range-finding sensor as the origin; setting a plurality of model regions including all of the distance measurement point clouds on the two-dimensional coordinate system or the three-dimensional coordinate system; On the two-dimensional coordinate system or the three-dimensional coordinate system, a region that is a blind spot from the distance measurement sensor due to the distance measurement point cloud is set as an occupied region; On the two-dimensional coordinate system or the three-dimensional coordinate system, an area other than the occupied area is set as an unoccupied area; selecting, from among the plurality of model regions, model regions in which the area or volume of an overlapping region where each of the model regions and the unoccupied region overlap is equal to or less than a predetermined value; The selected model region is recognized as the target object.

1. An object recognition method comprising:

2. The controller On the two-dimensional coordinate system or the three-dimensional coordinate system, the occupied area is set based on a ranging point among the ranging points that is the shortest distance from the ranging sensor in each direction from the ranging sensor.

2. The object recognition method according to claim 1.

3. The controller If there are a plurality of model regions whose overlapping areas have an area or volume equal to or less than the predetermined value, the model region whose overlapping area has the smallest area or volume is selected and recognized as the target object.

3. The object recognition method according to claim 1 or 2.

4. The controller If there is no model region where the area or volume of the overlapping region is equal to or less than the predetermined value, the distance measurement point group is divided, and distance measurement points on the surface of the same target object are re-extracted as the distance measurement point group from the divided distance measurement point group.

3. The object recognition method according to claim 1 or 2.

5. The controller If there are a plurality of model regions in which the area or volume of the overlapping region is equal to or less than the predetermined value, the model region in which the angular difference between the model region selected when the target object was recognized one frame before is equal to or less than the predetermined angular difference is selected, and the model region is recognized as the target object.

3. The object recognition method according to claim 1 or 2.

6. a distance measurement sensor mounted on a moving body and generating distance measurement point data relating to distance measurement points on the surface of a target object present around the moving body; a controller that recognizes the target object based on the distance measurement point data acquired from the distance measurement sensor. The controller extracting a plurality of pieces of distance measurement point data on the surface of the same target object from the distance measurement point data as a distance measurement point group; converting the range-finding point cloud into two-dimensional or three-dimensional coordinates with the position of the range-finding sensor as the origin; setting a plurality of model regions including all of the distance measurement point clouds on the two-dimensional coordinate system or the three-dimensional coordinate system; On the two-dimensional coordinate system or the three-dimensional coordinate system, a region that is a blind spot from the distance measurement sensor due to the distance measurement point cloud is set as an occupied region; On the two-dimensional coordinate system or the three-dimensional coordinate system, an area other than the occupied area is set as an unoccupied area; selecting, from among the plurality of model regions, model regions in which the area or volume of an overlapping region where each of the model regions and the unoccupied region overlap is equal to or less than a predetermined value; The selected model region is recognized as the target object. An object recognition device characterized by:

Images