Object recognition method and object recognition device

Abstract

This object recognition method and this object recognition device: generate, on the basis of distance measurement point data pertaining to a plurality of distance measurement points acquired by a sensor, a point group by clustering the plurality of distance measurement points; set sections having different positions in the vertical direction; and calculate, for each of the sections, an index value indicating the degree of spread of the point group in the horizontal direction in the section. The method and the device: extract, as a target section, a section in which an index value at a first time and an index value at a second time, which is a prescribed time in the past from the first time, are both less than a prescribed threshold; calculate a reference point on the basis of distance measurement points constituting a point group in the target section; and track the reference point to calculate object information of an object related to the point group.

Description

Object recognition method and object recognition device

[0001] The present disclosure relates to an object recognition method and an object recognition device.

[0002] A technology has been proposed in which position information for each of a plurality of detection points around a sensor is acquired, the plurality of detection points are classified into a group of detection points that constitute an object, a predetermined shape model is applied to the group of detection points based on the position information for each of the detection points that constitute the classified group of detection points, and a target point of the object is identified based on the shape model (see Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2017-215161

[0004] The technology described in Patent Document 1 has the problem that when the shape of an object, such as a pedestrian, changes, the target point cannot be stably identified and object information (position, speed, posture, etc.) cannot be calculated accurately.

[0005] The present disclosure has been made in view of the above-mentioned problems, and an object recognition method and an object recognition device are capable of calculating object information with high accuracy even when the shape of an object, such as a pedestrian, changes.

[0006] To solve the above-mentioned problems, an object recognition method and an object recognition device according to one aspect of the present disclosure cluster multiple ranging points based on ranging point data relating to multiple ranging points acquired by a sensor to generate a point cloud, set segments having different vertical positions, and calculate, for each segment, an index value indicating the degree of horizontal spread of the point cloud in the segment. A segment for which the index value at a first time and the index value at a second time a predetermined time before the first time are both less than a predetermined threshold is extracted as a target segment, a reference point is calculated based on the ranging points that constitute the point cloud in the target segment, and the reference point is tracked to calculate object information of an object related to the point cloud.

[0007] According to the present disclosure, even when the shape of an object such as a pedestrian changes, object information can be calculated with high accuracy.

[0008] Fig. 1 is a block diagram showing a configuration of an object recognition device according to an embodiment of the present disclosure. Fig. 2 is a flowchart showing processing of the object recognition device according to an embodiment of the present disclosure. Fig. 3 is a first example showing index values ​​corresponding to the shape of an object. Fig. 4 is a second example showing index values ​​corresponding to the shape of an object.

[0009] Next, embodiments of the present disclosure will be described in detail with reference to the drawings. In the description, the same components are designated by the same reference numerals and redundant description will be omitted.

[0010] [Configuration of Object Recognition Device] An example configuration of an object recognition device 1 will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of an object recognition device according to this embodiment. As shown in Fig. 1, the object recognition device 1 includes a distance measurement sensor 10 (sensor) and a controller 20.

[0011] The object recognition device 1 may be mounted on a vehicle with an automatic driving function, or on a vehicle without an automatic driving function. Furthermore, the object recognition device 1 may be mounted on a vehicle capable of switching between automatic driving and manual driving. Furthermore, the automatic driving function may be a driving assistance function that automatically controls only some of the vehicle control functions, such as steering control, braking force control, and driving force control, to assist the driver in driving. In this embodiment, the object recognition device 1 is described as being mounted on a vehicle with an automatic driving function.

[0012] 1, the object recognition device 1 may control various actuators such as a steering actuator, an accelerator pedal actuator, and a brake actuator based on the recognition results (position, shape, attitude, etc. of the object), thereby realizing highly accurate autonomous driving.

[0013] The distance measurement sensor 10 generates distance measurement point data relating to a plurality of distance measurement points.

[0014] For example, the distance measurement sensor 10 may include a sensor that measures the distance and direction to an object around the vehicle by emitting electromagnetic waves from an emission point around the vehicle and detecting the position of the reflection point based on the reflected wave of the emitted electromagnetic wave. One example of such a sensor is a LIDAR (Laser Imaging Detection and Ranging). A LIDAR is a sensor that emits light (laser light) from an emission point around a predetermined range around the vehicle, detects the position of a reflection point, which is a ranging point, based on the reflected wave, and generates ranging point data related to the ranging point.

[0015] Lidar measures the distance and direction to an object and recognizes the shape of the object by measuring the time it takes for the light (reflected wave) to bounce back after being emitted. Lidar can also obtain the positional relationship of objects in three dimensions. Mapping is also possible using the intensity of the reflected wave.

[0016] For example, the lidar scans the surroundings of the vehicle in the main scanning direction and the sub-scanning direction by changing the light irradiation direction. In particular, the lidar acquires range measurement point data for range measurement points located on the sides of a cone with the apex at the emission point by changing the angle from the horizontal plane (the horizontal plane in the lidar coordinate system) when emitting electromagnetic waves from the emission point. One side of the cone corresponds to one angle when the lidar emits electromagnetic waves.

[0017] For example, the lidar acquires ranging point data for ranging points included on the side surface of a single cone by scanning the side surface of the single cone (scanning in the main scanning direction).The lidar then acquires ranging point data for the side surfaces of multiple cones by scanning the side surface of the single cone (scanning in the sub-scanning direction) while changing the angle from the horizontal plane when emitting electromagnetic waves from the emission point.In this way, the lidar sequentially irradiates multiple ranging points around the vehicle with light.The lidar emits electromagnetic waves and generates ranging point data for ranging points located on the surface of the object based on the reflected waves of the electromagnetic waves from the surface of the object.

[0018] The LIDAR repeatedly emits light to all of the distance measurement points around the vehicle at predetermined time intervals. The LIDAR generates information (distance measurement point information) for each distance measurement point obtained by emitting light. The LIDAR then outputs the distance measurement point data to the controller 20.

[0019] The ranging point information includes position information of the ranging point. The position information is information indicating the position coordinates of the ranging point. The position coordinates may use a polar coordinate system represented by the direction from the lidar to the ranging point (yaw angle, pitch angle) and the distance from the lidar to the ranging point (depth). The position coordinates may use a three-dimensional coordinate system represented by x, y, and z coordinates with the installation position of the lidar as the origin. For example, the x and y coordinates may be coordinates on a horizontal plane in the lidar coordinate system, and the z coordinate may be a coordinate on an axis in the height direction perpendicular to the horizontal plane. Note that the x and y coordinates may be coordinates on a plane parallel to the road surface on which the vehicle is traveling, and the z coordinate may be a coordinate on an axis in the height direction above the road surface.

[0020] The ranging point information may also include time information of the ranging point. The time information is information indicating the time when the position information of the ranging point was generated (when the reflected electromagnetic wave was received). Additionally, the ranging point information may also include information on the intensity of the reflected wave from the ranging point (intensity information).

[0021] Alternatively, the distance measurement sensor 10 may be a stereo camera, which generates distance measurement point data relating to the positions of the distance measurement points using the principles of trigonometry based on a plurality of image data captured by the stereo camera.

[0022] The controller 20 processes the ranging point data generated by the ranging sensor 10. For example, the controller 20 is a general-purpose computer equipped with a CPU (Central Processing Unit), memory, and input / output units. A computer program for causing the computer to function as the object recognition device 1 is installed in the computer. By executing the computer program, the computer functions as multiple information processing circuits equipped in the object recognition device 1.

[0023] 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.

[0024] The controller 20 includes, as examples of a plurality of information processing circuits (information processing functions), a point cloud acquisition unit 21, an index value calculation unit 23, a segment extraction unit 25, a reference point calculation unit 27, an object information calculation unit 29, and an output unit 31. The controller 20 may also be expressed as an ECU (Electronic Control Unit).

[0025] The point cloud acquisition unit 21 generates a point cloud by clustering multiple ranging points based on the ranging point data. Each point cloud is a group of ranging points related to one or more ranging point data. For example, the point cloud acquisition unit 21 may simply treat multiple ranging points whose distances between them are equal to or less than a predetermined threshold as belonging to one point cloud and perform clustering.

[0026] Alternatively, the point cloud acquisition unit 21 may generate a point cloud by clustering a plurality of ranging points using various clustering algorithms such as k-means, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), or HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise).

[0027] The index value calculation unit 23 sets sections at different positions in the vertical direction and calculates an index value indicating the horizontal extent of the point cloud in each section. Here, a "section" refers to an area obtained by dividing a space vertically at a predetermined height. The "predetermined height" that serves as the basis for defining the sections may be set to a fixed value in advance or may be set according to the height of the point cloud. For example, the "predetermined height" may be set to the height obtained by dividing the height of the point cloud by a predetermined number so that the point cloud is divided into a predetermined number of areas in the vertical direction.

[0028] The index value calculation unit 23 extracts the distance measurement points included in each section from among the distance measurement points included in the point cloud, and calculates an index value indicating the spread for the extracted distance measurement points.

[0029] For example, the index value calculation unit 23 may set the variance of the horizontal positions of the ranging points for each section and each point group as the index value. Alternatively, the index value calculation unit 23 may set the maximum value of the horizontal distance of the ranging points for each section and each point group as the index value.

[0030] The index value calculation unit 23 may use the horizontal angle (apparent size) at which multiple ranging points are viewed from the ranging sensor 10 as an index value for each section and each point cloud. The larger the variance of the horizontal positions of the ranging points or the maximum value of the horizontal distance of the ranging points, the larger the apparent size. Also, the longer the distance from the ranging sensor 10 to the point cloud, the smaller the apparent size. Even if the apparent size is used instead of the variance of the horizontal positions of the ranging points or the maximum value of the horizontal distance of the ranging points as the index value calculated by the index value calculation unit 23, this does not affect the subsequent processing by the section extraction unit 25.

[0031] When processing is performed based on apparent size, using position information based on a polar coordinate system as distance measurement point data eliminates the need for conversion from polar coordinates to a three-dimensional coordinate system, thereby reducing calculation costs.

[0032] The index value calculation unit 23 may calculate the index value at a predetermined cycle, or may calculate the index value at a plurality of predetermined timings. In the following description, it is assumed that the index value calculation unit 23 calculates the index value at a first time. It is also assumed that the index value calculation unit 23 calculates the index value at a second time that is a predetermined time before the first time.

[0033] The segment extraction unit 25 extracts, as a target segment, a segment for which the index value at a first time and the index value at a second time that is a predetermined time in the past from the first time are both less than a predetermined threshold. More specifically, the segment extraction unit 25 acquires the index values ​​calculated by the index value calculation unit 23 for each segment and for each point cloud. Then, the segment extraction unit 25 extracts, for each point cloud, a segment for which the index value at the first time and the index value at the second time are both less than a predetermined threshold.

[0034] Here, the predetermined threshold may be determined in advance according to the characteristics of the object targeted by the object recognition device 1. Furthermore, the predetermined threshold may be set smaller as the distance from the distance measuring sensor 10 to the point cloud increases.

[0035] When the index value is the variance of the horizontal positions of the ranging points or the maximum value of the horizontal distances of the ranging points, the predetermined threshold may be set based on the characteristic length of the object targeted by the object recognition device 1. For example, when the object is a pedestrian, the predetermined threshold may be set based on the waist circumference or head width of an average pedestrian.

[0036] When the index value is the horizontal angle (apparent size) looking into multiple ranging points, the specified threshold may be set based on the value obtained by dividing the characteristic length of the object targeted by the object recognition device 1 by the characteristic length of the object targeted by the object recognition device 1.

[0037] When processing is performed based on apparent size, using position information based on a polar coordinate system as distance measurement point data eliminates the need for conversion from polar coordinates to a three-dimensional coordinate system, thereby reducing calculation costs.

[0038] Various methods for setting the predetermined threshold are applicable, and the method for setting the predetermined threshold is not limited to the example given here, as long as the object recognition device 1 can extract the target object.

[0039] The reference point calculation unit 27 calculates a reference point based on the ranging points that make up the point cloud in the target segment. For example, the reference point calculation unit 27 may calculate the center of gravity of the ranging points included in the target segment as the reference point for the point cloud. Alternatively, the reference point calculation unit 27 may apply a predetermined shape model to fit the ranging points included in the target segment and calculate a reference point based on the shape model. For example, various shape models such as a rectangular parallelepiped or a cylinder can be applied. The method of calculating a reference point based on the ranging points included in the target segment is not limited to the example given here.

[0040] The reference point calculation unit 27 calculates at least one reference point for each target segment. The reference point calculation unit 27 may calculate a plurality of reference points for each target segment.

[0041] Alternatively, if there are multiple target segments extracted by the segment extraction unit 25, the reference point calculation unit 27 may calculate the reference point based on the ranging points that make up the point cloud in the target segment with the highest position. A ranging point that is located at a high position on the object is less likely to be blocked by other objects, making it easier to stably track the reference point.

[0042] Furthermore, if there are multiple target segments extracted by the segment extraction unit 25, the reference point calculation unit 27 may calculate the reference point based on the ranging points constituting the point cloud in the target segment in which the absolute value of the change in index value from the second time to the first time is smallest. This reduces the influence of changes in the shape of the object on the position of the reference point, making it easier to stabilize the position of the reference point being tracked across multiple times. As a result, it becomes possible to calculate object information with high accuracy.

[0043] In addition, if there are multiple target segments extracted by the segment extraction unit 25, the reference point calculation unit 27 may calculate the reference point based on the ranging point that constitutes the point cloud in the target segment having the highest position, and / or the ranging point that constitutes the point cloud in the target segment having the smallest absolute value of the change in index value from the second time to the first time.

[0044] The object information calculation unit 29 calculates object information of the object related to the point cloud by tracking the reference points calculated by the reference point calculation unit 27. Here, the object information is information related to at least one of the position, velocity, and orientation of the object.

[0045] For example, the object information calculation section 29 may track the position of the object by tracking the reference point calculated by the reference point calculation section 27 .

[0046] Furthermore, the object information calculation unit 29 may predict current object information based on previously calculated object information. For example, the object information calculation unit 29 may predict the current object position based on previously calculated object positions and velocities. Here, when predicting the object position, it may be assumed that the object continues moving at the same velocity as previously calculated.

[0047] The object information calculation unit 29 may determine the current object information by calculating a weighted average of the predicted current object information and the calculated current object information. In this case, the object information calculation unit 29 may set a smaller weight for the calculated current object information as the index value for the target section increases. In other words, the object information calculation unit 29 may treat the calculated current object information as being less accurate as the index value for the target section increases. In this case, the predicted current object information is given priority for use in determining the object information as the index value for the target section increases.

[0048] More specifically, the object information calculation unit 29 may use the index value in the target section as the reliability of the observation. To use the index value as the reliability of the observation, the object information calculation unit 29 may determine a parameter representing the variance of the probability distribution of the observation in various filters such as a Kalman filter or a particle filter, based on the index value in the target section. The greater the horizontal spread of the point cloud, the greater the variance of the probability distribution of the observation, allowing for estimation of object information at the current time with greater emphasis on prediction.

[0049] For example, if the point cloud is large, the uncertainty of the object position is considered to be high, and the accuracy of determining the object information may decrease. However, by setting a small weight for the calculated current object information, the object information can be determined based on the predicted current object information, and the accuracy of determining the object information can be prevented from decreasing.

[0050] Alternatively, the object information calculation unit 29 may predict the position of the current reference point based on object information calculated in the past, and may associate the object associated with the predicted reference point with the object associated with the calculated reference point based on the predicted and calculated positions of the reference point. More specifically, if the predicted position of the reference point and the calculated position of the reference point are close (within a predetermined distance), the object associated with the predicted reference point and the object associated with the calculated reference point may be treated as the same object.

[0051] By tracking a reference point that allows stable tracking, it is possible to link the object information obtained by prediction with the object information obtained based on the observation results from the sensor, thereby enabling stable tracking of the object.

[0052] The output unit 31 outputs data relating to the object information calculated by the object information calculation unit 29. Alternatively, the output unit 31 may output the reference points calculated by the reference point calculation unit 27 or data representing a shape model.

[0053] [Processing Procedure of Object Recognition Apparatus] Next, a processing procedure of the object recognition apparatus 1 according to this embodiment will be described with reference to the flowchart of Fig. 2. Fig. 2 is a flowchart showing the processing of the object recognition apparatus according to this embodiment. The processing of the object recognition apparatus 1 shown in Fig. 2 may be repeatedly executed at a predetermined cycle.

[0054] In step S101, the distance measurement sensor 10 generates distance measurement point data relating to a plurality of distance measurement points.

[0055] In step S103, the point cloud acquisition unit 21 generates a point cloud by clustering the plurality of distance measurement points based on the distance measurement point data.

[0056] In step S105, the index value calculation unit 23 sets sections having different positions in the vertical direction, and calculates an index value for each section that indicates the degree of spread of the point cloud in the horizontal direction in the section.

[0057] Fig. 3 is a first example showing index values ​​corresponding to the shape of an object. In particular, Fig. 3 shows a state in which a pedestrian PD (object) has taken a step and has an open stride. Fig. 4 is a second example showing index values ​​corresponding to the shape of an object. In particular, Fig. 4 shows a state in which a pedestrian PD (object) has taken a closed stride, just before taking a step.

[0058] As shown in FIGS. 3 and 4, if the distribution of distance measurement points is spread depending on the shape of the object, the index value will be large, and if the distribution of distance measurement points is not spread, the index value will be small.

[0059] In step S107, the segment extraction unit 25 extracts segments whose index values ​​are less than a predetermined threshold as target segments. The predetermined threshold TH is shown in Figures 3 and 4. In the example shown in Figure 3, segment R8 is extracted as the target segment. In the example shown in Figure 4, segments R1 and R8 are extracted as target segments.

[0060] In step S109, the reference point calculation unit 27 calculates a reference point based on the ranging points that make up the point cloud in the target section. For example, as shown in Fig. 4, if there are multiple target sections, the reference point calculation unit 27 may calculate a reference point based on the ranging points that make up the point cloud in the target section with the highest position. In the example shown in Fig. 4, sections R1 and R8 are extracted as target sections, and the reference point calculation unit 27 calculates a reference point based on the ranging points that make up the point cloud in section R8 with the highest position.

[0061] In step S111, the object information calculation unit 29 tracks the reference points calculated by the reference point calculation unit 27 to calculate object information of the objects related to the point cloud.

[0062] In step S113 , the output unit 31 outputs data relating to the object information calculated by the object information calculation unit 29 .

[0063] [Effects of the Embodiment] As described in detail above, the object recognition method and object recognition device according to this embodiment generate a point cloud by clustering multiple ranging points based on ranging point data related to multiple ranging points acquired by a sensor, set segments having different vertical positions, and calculate, for each segment, an index value indicating the degree of horizontal spread of the point cloud in the segment. A segment for which the index value at a first time and the index value at a second time a predetermined time before the first time are both less than a predetermined threshold is extracted as a target segment, a reference point is calculated based on the ranging points that make up the point cloud in the target segment, and the reference point is tracked to calculate object information of an object related to the point cloud.

[0064] This allows for accurate calculation of object information even when the shape of an object, such as a pedestrian, changes. In particular, the influence of changes in the shape of the object on the position of the reference point is reduced, making it easier to stabilize the position of the reference point being tracked across multiple times. As a result, object information can be calculated with high accuracy.

[0065] In the object recognition method and object recognition device according to this embodiment, the object information may be at least one of the position, velocity, and orientation of the object, which allows the position, velocity, orientation, etc. of the object to be calculated with high accuracy even when the shape of the object changes.

[0066] Furthermore, the object recognition method and object recognition device according to this embodiment may calculate the reference point based on the ranging point that constitutes the point cloud in the target section having the highest position. The ranging point at a high position on the object is less likely to be blocked by other objects, making it easier to stably track the reference point.

[0067] Furthermore, the object recognition method and object recognition device according to this embodiment may calculate the reference point based on the ranging points constituting the point cloud in the target section in which the absolute value of the change in index value from the second time to the first time is smallest. This reduces the influence of changes in the shape of the object on the position of the reference point, making it easier to stabilize the position of the reference point tracked across multiple times. As a result, it becomes possible to calculate object information with high accuracy.

[0068] Furthermore, the object recognition method and object recognition device according to this embodiment may predict current object information based on previously calculated object information, and determine the current object information by weighted averaging the predicted current object information and the calculated current object information. The larger the index value in the target category, the smaller the weighting of the calculated current object information. This allows the likelihood of the calculated object information to be adjusted according to the spread of the point cloud. As a result, it is possible to continuously determine object information even when the spread of the point cloud is large.

[0069] For example, if the point cloud is large, the uncertainty of the object position is considered to be high, and the accuracy of determining the object information may decrease. However, by setting a small weight for the calculated current object information, the object information can be determined based on the predicted current object information, and the accuracy of determining the object information can be prevented from decreasing.

[0070] Furthermore, the object recognition method and object recognition device according to this embodiment may predict the position of a current reference point based on object information calculated in the past, and associate the object associated with the predicted reference point with the object associated with the calculated reference point based on the predicted position of the reference point and the calculated position of the reference point. In this way, by tracking a reference point that can be stably tracked, it is possible to link object information obtained by prediction with object information obtained based on observation results by a sensor. As a result, it becomes possible to stably track objects.

[0071] Each of the functions described in the above embodiments may be implemented by one or more processing circuits, including programmed processors, electrical circuits, and even devices such as application specific integrated circuits (ASICs), or circuit components arranged to perform the described functions.

[0072] Although the contents of the present disclosure have been described above based on the embodiments, the present disclosure is not limited to these descriptions, and various modifications and improvements are possible, which will be apparent to those skilled in the art. The descriptions and drawings that form part of this disclosure should not be understood as limiting the present disclosure. Various alternative embodiments, examples, and operating techniques will be apparent to those skilled in the art from this disclosure.

[0073] Of course, the present disclosure includes various embodiments not described herein. Therefore, the technical scope of the present disclosure is defined only by the invention-specifying matters according to the scope of the claims that are appropriate from the above description.

[0074] REFERENCE SIGNS LIST 1 Object recognition device 10 Distance measurement sensor 20 Controller 21 Point cloud acquisition unit 23 Index value calculation unit 25 Classification extraction unit 27 Reference point calculation unit 29 Object information calculation unit 31 Output unit

Claims

1. An object recognition method for an object recognition device including a sensor that generates distance measurement point data for a plurality of distance measurement points and a controller that processes the distance measurement point data, wherein the controller: clusters the plurality of distance measurement points based on the distance measurement point data to generate a point cloud; sets sections having different positions in the vertical direction; calculates an index value indicating the degree of spread of the point cloud in the horizontal direction for each of the sections; extracts, as a target section, a section in which the index value at a first time and the index value at a second time, which is a predetermined time in the past from the first time, are both less than a predetermined threshold value; calculates a reference point based on the distance measurement points constituting the point cloud in the target section; and calculates object information of an object related to the point cloud by tracking the reference point.

2. The object recognition method according to claim 1, wherein the object information is at least any one of information on the position, velocity, and attitude of the object.

3. The object recognition method according to claim 1 or 2, wherein the controller calculates the reference point based on the distance measurement points constituting the point cloud in the target section having the highest position.

4. The object recognition method according to any one of claims 1 to 3, wherein the controller calculates the reference point based on the distance measurement points constituting the point cloud in the target section in which the absolute value of the change in the index value from the second time to the first time is the smallest.

5. The object recognition method according to any one of claims 1 to 4, wherein the controller predicts current object information based on the object information calculated in the past, and when determining the current object information by weighted-averaging the predicted current object information and the calculated current object information, the greater the index value in the target section, the smaller the weight assigned to the calculated current object information.

6. The controller predicts the position of the current reference point based on the object information calculated in the past, and associates the object related to the predicted reference point with the object related to the calculated reference point based on the predicted position of the reference point and the calculated position of the reference point. The object recognition method according to any one of claims 1 to 5, characterized in that.

7. An object recognition device comprising a sensor that generates distance measurement point data for a plurality of distance measurement points and a controller that processes the distance measurement point data, wherein the controller clusters the plurality of distance measurement points based on the distance measurement point data to generate a point cloud, sets sections having different positions in the vertical direction, calculates an index value indicating the degree of spread of the point cloud in the horizontal direction for each section, extracts as a target section a section in which the index value at a first time and the index value at a second time that is a predetermined time in the past from the first time are both less than a predetermined threshold value, calculates a reference point based on the distance measurement points constituting the point cloud in the target section, and calculates object information of an object related to the point cloud by tracking the reference point. An object recognition device characterized by that.

Images