JP7673437B2 - OBJECT RECOGNITION DEVICE AND OBJECT RECOGNITION METHOD - Google Patents

Description

This disclosure relates to technology for recognizing objects.

There is known an object recognition device that recognizes objects around a vehicle using the detection results of a LiDAR mounted on the vehicle. For example, Patent Document 1 discloses a technology that dynamically changes a threshold value when creating clusters to create multiple cluster candidates, compares the created cluster candidates with clusters created in the past, and recognizes the cluster with the closest shape combination as a single object.

JP 2013-228259 A

However, the technology described in Patent Document 1 requires the history of clusters that have been created in the past. As a result, there is no cluster history at the time of the first detection, and there is a risk that the accuracy of object detection will decrease. Therefore, there is a need for a technology in an object recognition device that can accurately distinguish between multiple objects that exist close to each other.

According to an embodiment of the present disclosure, an object recognition device (10) is provided. The object recognition device includes a detection point cloud acquisition unit (22) that acquires a detection point cloud composed of a plurality of detection points (KP) that represent objects (Ob1, Ob2, Ob3, Ob4) that exist around the object recognition device, a contour point cloud acquisition unit (23) that extracts contour points (RP) of the object from the detection point cloud to acquire a contour point cloud, a representative point cloud acquisition unit (24) that acquires a representative point cloud by extracting a representative point (DP) from the contour point cloud, and a first vector acquisition unit (25) that acquires a vector between the first point and the second point by setting three points selected from the representative point cloud as a first point (P1), a second point (P2), and a third point (P3). and an inverse convexity determination unit (25) that determines whether or not the contour point group includes an inverse convex shape (ITs) using an inter-vector angle (θ) that is the angle between a vector (V1) and a second vector (V2) that is a vector between the second point and the third point, and an object recognition unit (26) that, when it is determined that the contour point group includes the inverse convex shape, separates the detection point group into a first detection point group (G1) that represents a wall, which is a linear object among the objects that exist around the object recognition device, and a second detection point group (G2) that represents a different object that is different from the wall and exists in the vicinity of the wall.

According to this form of object recognition device, three points selected from the representative point group are designated as the first point, the second point, and the third point, and the vector angle between the first vector, which is the vector between the first point and the second point, and the second vector, which is the vector between the second point and the third point, is used to determine whether or not the contour point group contains an inverted convex shape. If it is determined that an inverted convex shape is contained, it is recognized that a linear object and an object other than a linear object exist close to each other, and therefore multiple objects that exist close to each other can be accurately distinguished.

The present disclosure can be realized in various forms. For example, it can be realized in the form of an object recognition method, a distance measuring device, a distance measuring method, a computer program for realizing these devices and methods, a storage medium storing such a computer program, etc.

1 is an explanatory diagram showing a schematic configuration of a vehicle equipped with an object recognition device according to an embodiment of the present disclosure. FIG. 1 is a block diagram showing a functional configuration of an object recognition device. FIG. 1 is an explanatory diagram illustrating an example of a scene in which an object recognition process is executed. FIG. 13 is an explanatory diagram for explaining an inverted convex shape. FIG. 11 is a flowchart showing the procedure of an object recognition process. FIG. 4 is an explanatory diagram illustrating a schematic diagram of how a contour point group is acquired. FIG. 11 is an explanatory diagram illustrating an example of a method for obtaining a representative point. FIG. 11 is an explanatory diagram illustrating a schematic diagram of another example of a method for obtaining a representative point. 11 is a flowchart showing the processing procedure of a representative point group acquisition process. FIG. 4 is an explanatory diagram illustrating a schematic diagram of how a representative point group is acquired. 13 is a flowchart showing the procedure of a reverse convexity determination process. FIG. 11 is an explanatory diagram illustrating a schematic diagram of a reverse convexity determination process being executed. FIG. 4 is an explanatory diagram for explaining an angle between vectors. 11 is a flowchart showing the procedure of a separation line determination process. FIG. 11 is an explanatory diagram illustrating a schematic view of how a separation line determination process is executed. FIG. 11 is an explanatory diagram illustrating a schematic diagram of another example of a method for obtaining a representative point.

A1. Embodiment:
An object recognition device and an object recognition method will be described below. As shown in FIG. 1, an object recognition device 10 of this embodiment is mounted on a vehicle 100 and detects objects present in the vicinity in front of the vehicle 100, such as other vehicles, pedestrians, and buildings. In this embodiment, the object recognition device 10 is configured by LiDAR (Light Detection And Ranging). The object recognition device 10 irradiates irradiation light Lz and receives reflected light from an object. In FIG. 1, the emission center position of the irradiation light Lz is the origin, the forward direction of the vehicle 100 is the Y axis, the direction from left to right in the width direction of the vehicle 100 passing through the origin is the X axis, and vertical information passing through the origin is the Z axis.

The illumination light Lz is elongated in the Z-axis direction, and is illuminated over a predetermined vertical range Ar with each illumination. The illumination light Lz is also illuminated over the entire predetermined measurement range MR by one-dimensional scanning in the horizontal direction. The illumination light Lz is illuminated in the measurement range MR at an angle that corresponds to the resolution of the object recognition device 10.

The object recognition device 10 acquires a detection point group consisting of a plurality of detection points that represent an object identified by the reflected light of the irradiation light Lz in the measurement range MR. The detection points are points that indicate positions where at least a part of the object identified by the reflected light may be present. The object recognition device 10 also detects the distance to the object by measuring the time from when the irradiation light Lz is emitted to when the reflected light is received, that is, the time of flight of light TOF (Time of Flight). The object recognition device 10 performs clustering on the acquired detection point group and classifies the detection point group into a plurality of clusters, thereby classifying the object.

As shown in FIG. 2, the object recognition device 10 includes a CPU 20, a memory 30, and an input/output interface 11. The CPU 20, memory 30, and input/output interface 11 are connected via a bus 15 to enable bidirectional communication. The memory 30 includes a ROM, a RAM, and an EEPROM. The light emission control unit 41 and the light reception control unit 51 are each connected to the input/output interface 11 via a control signal line. A light emission control signal is sent to the light emission control unit 41, a light reception control signal is sent to the light reception control unit 51, and an incident light intensity signal is received from the light reception control unit 51.

The light emission control unit 41 drives the light emitting element 42 to emit the irradiation light Lz at a timing according to the light emission control signal input from the control unit 21 via the input/output interface 11. The light emitting element 42 is, for example, an infrared laser diode, and emits infrared laser light as the irradiation light Lz. There may be one or more light emitting elements 42.

The light receiving control unit 51 receives a light receiving control signal from the control unit 21 via the input/output interface 11, which instructs the light receiving process for object detection, and outputs an incident light intensity signal indicating the amount of incident light or the intensity of the incident light incident on the light receiving element 52. The light receiving element 52 has a configuration in which multiple pixels are arranged vertically and horizontally on a flat optical sensor, and is composed of, for example, a SPAD (Single Photon Avalanche Diode) or other photodiode. The light receiving control unit 51 converts a current generated according to the amount of incident light or the intensity of the incident light incident on the light receiving element 52 into a voltage, and outputs, for example, a pixel value indicating a luminance value or brightness to the control unit 21 as an incident light intensity signal.

The CPU 20 deploys and executes the programs stored in the memory 30 to function as a control unit 21, a detection point cloud acquisition unit 22, a contour point cloud acquisition unit 23, a representative point cloud acquisition unit 24, a reverse convexity determination unit 25, an object recognition unit 26, and a separation line determination unit 27.

The control unit 21 controls the operation of the entire object recognition device 10.

The detection point group acquisition unit 22 acquires detection points or detection point groups indicating one or more representative positions of an object from pixel values indicated by the incident light intensity signal. The detection points are acquired as three-dimensional coordinates (X, Y, Z). The detection points may be expressed as two-dimensional coordinates (X, Y) instead of three-dimensional coordinates. The contour point group acquisition unit 23 extracts contour points of the object from the detection point group to acquire a contour point group consisting of multiple contour points. The representative point group acquisition unit 24 extracts representative points from the contour point group to acquire a representative point group consisting of multiple representative points. The reverse convexity determination unit 25 selects three points from the representative point group and determines whether or not the vector connecting these three points has a reverse convex shape. The reverse convex shape will be described later. The object recognition unit 26 performs clustering processing on the detection point group to recognize an object from multiple detection points. The separation line determination unit 27 determines a line (hereinafter referred to as a "separation line") for separating the detection points into a detection point group that constitutes one object and a detection point group that constitutes another object.

A2. Object Recognition Processing:
In this embodiment, the processing procedure of the object recognition processing will be described assuming that the object recognition processing is being executed in the driving scene shown in Fig. 3. As shown in Fig. 3, the vehicle 100 is running in the forward direction on the road RD. The road RD is provided with a sidewalk RS on which a linear wall Ob1 is installed, and a pedestrian Ob3 is walking in the forward direction along the wall Ob1. Trees Ob2 and Ob4 are planted on the sidewalk RS. The object recognition device 10 executes the object recognition processing described later, and can detect objects Ob1, Ob2, Ob3, and Ob4 that exist around the vehicle 100 in the forward direction.

Figure 4 shows a schematic example of a bird's-eye view of detection point group CL1 that can be acquired in the driving scene shown in Figure 3 from vertically above (+Z direction). Figure 4 shows an enlarged view of the vicinity of pedestrian Ob3. For ease of explanation, detection points KP representing pedestrian Ob3 are hatched in Figure 4. Detection point group CL1 is made up of a first detection point group G1 and a second detection point group G2. The first detection point group G1 is a collection of detection points KP corresponding to walls Ob1 and trees Ob2, and the second detection point group G2 is a collection of detection points KP corresponding to pedestrian Ob3.

As shown in FIG. 3, the wall Ob1 and the pedestrian Ob3 are close to each other. In order to recognize the pedestrian Ob3 as an object different from the wall Ob1, it is necessary to accurately separate the second detection point group G2 from the first detection point group G1. Here, the inventors of the present disclosure have found that when a linear object, such as a wall and a person, and another object are close to each other, an inverted convex shape ITs is included in the contour line when the detection point group is viewed from above. Specifically, as shown in FIG. 4, an inverted convex shape ITs having a convex portion protruding toward the direction away from the object recognition device 10 in the traveling direction of the vehicle 100 appears near the boundary between the first detection point group and the second detection point group G2 on the contour line RL of the detection point group CL1. The inventors of the present disclosure have found that such an inverted convex shape ITs is detected only when the detection points of multiple objects are combined, that is, when multiple objects are close to each other.

Figure 5 shows detection point groups CL2 and CL3 when single objects Ob5 and Ob6 exist independently. A convex shape Ts having a convex portion protruding toward the object recognition device 10 appears in the detection point group CL2 representing object Ob5. Similarly, a convex shape Ts having a convex portion protruding toward the object recognition device 10 appears in the detection point group CL3 representing object Ob6. As can be seen by comparing Figures 4 and 5, the convex shape Ts and the inverted convex shape ITs have different protruding directions of the convex bend. The convex bend protrudes in a direction away from the object recognition device 10 in the inverted convex shape ITs, and protrudes toward the object recognition device 10 in the convex shape Ts.

In the object recognition process of this embodiment, the boundaries of nearby objects are identified by detecting the inverse convex shape ITs from the detection point cloud, and the detection point cloud is separated into detection point clouds for each object. Although a detailed explanation will be given later, when detecting the inverse convex shape ITs, the detection point cloud is not affected by objects with irregular contour shapes, such as trees Ob2 and Ob4. Note that in the following explanation, as shown in FIG. 3, a wall Ob1 parallel to the Y axis is used as a linear object, but the object recognition process of this embodiment can also be applied to cases where a linear object exists that intersects with the object recognition device 10, such as object Ob6 shown in FIG. 5.

The object recognition process shown in FIG. 6 is executed repeatedly at a predetermined time interval, for example, at any time in the range of 100 milliseconds to 200 milliseconds, from when the control system of the vehicle 100 is started to when it is stopped, or from when the start switch is turned on to when it is turned off. The detection point cloud acquisition unit 22 acquires a detection point cloud (step S10). The contour point cloud acquisition unit 23 extracts contour points of the object from the detection point cloud to acquire a contour point cloud (step S20).

Figure 7 shows how the contour point group is acquired from the detection point group. Figure 7 shows an example of a bird's-eye view of the detection point group from vertically above (+Z direction). In step S20, the contour point group acquisition unit 23 extracts the detection point KP with the shortest detection distance for each predetermined angle range in the horizontal direction as the contour point RP.

As shown in FIG. 6, the representative point group acquisition unit 24 extracts representative points from the contour point group to acquire the representative point group (step S30). Specifically, the representative point group acquisition unit 24 acquires the representative point group by thinning out the detection points KP from the contour point group in accordance with predetermined conditions. In other words, the representative point group is constructed by extracting the detection points KP that meet the predetermined conditions from the contour point group. In this embodiment, contour points whose distance between the contour points is equal to or greater than a predetermined distance are acquired as the representative points.

8 shows how the representative point group is acquired when the predetermined distance is set to a distance greater than a predetermined reference value. FIG. 9 shows how the representative point group is acquired when the predetermined distance is set to a distance smaller than a predetermined reference value. As shown in FIG. 8, when the predetermined distance is relatively large, the contour line RL1 obtained by sequentially connecting the representative points DP is smooth. Therefore, for example, the inverse convex shape ITs is not detected at positions PS1 and PS2, and the inverse convex shape ITs that should be detected cannot be detected. On the other hand, as shown in FIG. 9, when the predetermined distance is relatively small, the contour line RL2 obtained by sequentially connecting the representative points DP becomes jagged. Therefore, for example, the inverse convex shape ITs is erroneously detected at positions PS3 and PS4. Therefore, in this embodiment, as described later, the representative points are extracted while shifting the contour points one by one, and the inverse convex shape ITs is detected with high accuracy.

As shown in FIG. 10, the representative point group acquisition unit 24 sets a reference point (step S305). Specifically, the representative point group acquisition unit 24 sets each contour point constituting the contour point group as a reference point one by one in a predetermined order. For example, the representative point group acquisition unit 24 sets the reference points in order from the contour point RP that corresponds to the end point in the -Y direction in the contour point group toward the +Y direction.

The representative point group acquisition unit 24 acquires the contour points RP that are at a predetermined distance or more from the reference point as representative points DP (step S310). Specifically, the representative point group acquisition unit 24 sequentially identifies, for all contour points RP included in the contour point group, points that are at a predetermined distance or more from the reference point as representative points, and acquires a representative point group that is a collection of multiple representative points.

The representative point group acquisition unit 24 determines whether all of the contour points RP constituting the contour point group have been set as reference points (step S315). If it is determined that all of the contour points RP have not been set as reference points (step S315: NO), the representative point group acquisition unit 24 executes the above-mentioned step S305 to set the contour point RP adjacent to the contour point RP currently set as the reference point as the reference point. If it is determined in the above-mentioned step S315 that all of the contour points RP have been set as reference points (step S315: YES), the representative point group acquisition process ends.

Figure 11 shows how representative point groups are extracted. Representative point group 1 is a representative point group obtained when contour point rp1 is set as the reference point SP, representative point group 2 is a representative point group obtained when contour point rp2 adjacent to contour point rp1 is set as the reference point SP, and representative point group 3 is a representative point group obtained when contour point rp3 adjacent to contour point rp2 is set as the reference point SP.

As shown in representative point group 1, in step S310, first, the representative point group acquisition unit 24 extracts a contour point that is a predetermined distance or more from the contour point rp1, which is the reference point SP, as representative point Dp11. In this embodiment, the "predetermined distance d1" is, for example, 2 meters. Note that distance d1 may be any distance in the range of 0.5 meters to 5 meters instead of 2 meters. Next, the representative point group acquisition unit 24 extracts a contour point that is a distance d1 or more from the representative point Dp11 extracted immediately before as representative point Dp12. In the same manner, the representative point group acquisition unit 24 sequentially extracts contour points that are a distance d1 or more from the representative point extracted immediately before as representative points Dp13 and Dp14.

As shown in representative point group 2, when contour point rp2 is set as the reference point SP in step S305 and step S310 is executed, the representative point group acquisition unit 24 extracts a contour point that is a distance d1 or more from contour point rp2, which is the reference point SP, as representative point Dp21. Then, contour points that are a distance d1 or more from the most recently extracted representative point Dp21 are sequentially extracted as representative points Dp22, Dp23, and Dp24.

Similarly, in representative point group 3, a contour point that is a distance d1 or more from contour point rp3, which is the reference point SP, is extracted as representative point Dp31, and contour points that are a distance d1 or more from representative point Dp31 are extracted successively as representative points Dp31, Dp32, and Dp33. Note that when the contour point extracted as the representative point, for example, representative point Dp11 in representative point group 1, is set as the reference point SP, the combination of representative points that can be extracted will be the same as the representative point group extracted in the past. For this reason, after execution of step S305 and before execution of step S310 shown in FIG. 10, the representative point group acquisition unit 24 determines whether the combination of representative points that can be extracted will be the same as the representative point group extracted up to the previous time, and if it is determined that the combination will be the same, step S310 may be omitted.

As shown in FIG. 6, the reverse convexity determination unit 25 performs a reverse convexity determination (step S40).

As shown in FIG. 12, the reverse convex determination unit 25 sets an end point of the representative point group as the starting point (step S405). The reverse convex determination unit 25 selects three representative points (step S410). The reverse convex determination unit 25 calculates the inter-vector angle (step S415). Specifically, as shown in FIG. 14, the reverse convex determination unit 25 calculates the angle θ between the first vector V1, which is the vector between the first point P1 and the second point P2, and the second vector V2, which is the vector between the second point P2 and the third point P3 (hereinafter referred to as the "inter-vector angle"), using the following formula (1).

12, the reverse convex determination unit 25 determines whether the orientation is reverse convex (step S420). Specifically, the reverse convex determination unit 25 uses the sign of the calculated inter-vector angle to determine whether the first vector V1 and the second vector V2 form an inverse convex shape ITs. More specifically, the reverse convex determination unit 25 first calculates the angle θref between the first vector V1 and the third vector V3, which is the vector between the second point P2 and the object recognition device 10 (hereinafter referred to as the "reference inter-vector angle") using the following formula (2).

Next, the reverse convex determination unit 25 determines whether the sign of the inter-vector angle θ and the sign of the reference inter-vector angle θref are the same. If both signs are the same, the reverse convex determination unit 25 determines that the orientation is reverse convex (step S420: YES) and counts up the reverse convex level (step S425). In this embodiment, the "reverse convex level" means the degree of certainty that the representative point group includes the reverse convex shape ITs. The reverse convex determination unit 25 stores the three points P1, P2, and P3 selected in the above-mentioned step S410 in the memory 30 as reverse convex candidate points (step S430). The reverse convex determination unit 25 determines whether the inter-vector angle θ is equal to or greater than the determination threshold (step S435). In this embodiment, the determination threshold is, for example, 60 degrees. If it is determined that the inter-vector angle θ is equal to or greater than the determination threshold (step S435: YES), the reverse convex determination unit 25 sets the reverse convex determination flag to TRUE (step S440). The reverse convexity determination unit 25 stores the three points P1, P2, and P3 selected in the above-mentioned step S410 as reverse convex points in the memory 30 (step S445).

The reverse convexity determination unit 25 determines whether all representative points in the representative point group have been set as starting points (step S450). If it is determined that all representative points have not been set as starting points (step S450: NO), the reverse convexity determination unit 25 shifts the phase (step S475). Specifically, the reverse convexity determination unit 25 sets the first point P1, which is the starting point when selecting the three representative points in the above-mentioned step S410, to the representative point adjacent to the first point P1. Then, the above-mentioned steps S410 to S475 are executed.

If the sign of the inter-vector angle θ and the sign of the reference inter-vector angle θref are not the same in step S420, the reverse convex determination unit 25 determines that the direction is not reverse convex (step S420: NO), and the reverse convex determination unit 25 counts down the reverse convex level (step S455). The reverse convex determination unit 25 determines whether the reverse convex level is smaller than the reverse convex level lower limit (step S460). If it is determined that the reverse convex level is equal to or greater than the reverse convex level lower limit (step S460: NO), the above-mentioned step S475 is executed. On the other hand, if it is determined that the reverse convex level is smaller than the reverse convex level lower limit (step S460: YES), the reverse convex determination unit 25 sets the reverse convex determination flag to FALSE (step S465). The reverse convex determination unit 25 resets the reverse convex point (step S470). Specifically, the reverse convexity determination unit 25 erases the reverse convexity points stored in the memory 30 in the above-mentioned step S445. After execution of step S470, the above-mentioned step S450 is executed. If it is determined in the above-mentioned step S450 that all representative points have been set as starting points (step S450: YES), the reverse convexity determination process ends.

In phase 1 shown in FIG. 13, the representative point P1, which is an end point, is set as the starting point, the representative point P2, which is a predetermined distance d1 or more from the representative point P1, is selected as the second point, and the representative point P3, which is a predetermined distance d1 or more from the second point P2, is selected as the third point. An angle θ is calculated as the inter-vector angle between a first vector V1 connecting the first point P1 and the second point P2, and a second vector V2 connecting the second point P2 and the third point P3. In phase 2, the representative point adjacent to the starting point of phase 1 is set as the starting point, and three points P1, P2, and P3 are selected. In phase 2, the direction of the reverse convexity is opposite to that of phase 1. Therefore, the reverse convexity level is counted down. In phase 3, the starting point moves one more in the +Y direction, and three points P1, P2, and P3 are selected. In phase 3, the direction has changed to the reverse convexity, but the inter-vector angle θ is smaller than the definite threshold. Therefore, the three points P1, P2, and P3 selected in phase 3 are stored as reverse convex candidate points. In phase 4, the inter-vector angle θ is larger in the reverse convex direction. Therefore, the reverse convex level is counted up, and the selected three points P1, P2, and P3 are stored as reverse convex candidate points. In phase 5, compared to phase 4, the inter-vector angle θ is larger in the reverse convex direction and is equal to or greater than the definite threshold. Therefore, the selected three points P1, P2, and P3 are stored as reverse convex points. After that, the position of the starting point is moved by one point, and the calculation of the inter-vector angle θ, the determination of the reverse convex direction, and the determination of the inter-vector angle θ are performed.

As shown in FIG. 6, the reverse convex determination unit 25 determines whether or not an inverse convex shape ITs has been detected (step S50). If it is determined that an inverse convex shape ITs has been detected (step S50: YES), the separation line determination unit 27 determines a separation line (step S60). Specifically, the separation line determination unit 27 determines a separation line using the reverse convex points obtained by shifting the phase in the reverse convex determination process. More specifically, the separation line determination unit 27 determines a separation line using a group of reverse convex center points formed by the second point P2, which is the center point of the reverse convex points.

As shown in FIG. 15, the separation line determination unit 27 selects two points from the group of reverse convex center points (step S605). Specifically, the separation line determination unit 27 selects two points from each of the reverse convex center points constituting the group of reverse convex center points in a predetermined order. In the following description, the two points selected in step S605 are also referred to as "selected points."

Figure 16 shows how the separation line determination process is executed. For ease of explanation, in Figure 16, it is assumed that the inverse convex center point group is composed of points C1, C2, C3, C4, C5, C6, C7, C8, C9, C10, C11, and C12. The left side of Figure 16 shows the case where points C1 and C2 are selected points, the center of Figure 12 shows the case where points C1 and C3 are selected points, and the right side of Figure 12 shows the case where points C1 and C4 are selected points.

As shown in FIG. 15, the separation line determination unit 27 calculates a line passing through the two points selected in the above-mentioned step S605 (step S610). In the example shown in FIG. 16, the lines SL1, SL2, and SL3 are calculated.

As shown in FIG. 15, the separation line determination unit 27 calculates the residual of each inverse convex center point (step S615). Specifically, the separation line determination unit 27 calculates the difference between the regression equation representing the line calculated in step S610 and each of the inverse convex center points C1, C2, C3, C4, C5, C6, C7, C8, C9, C10, C11, and C12. The separation line determination unit 27 determines whether all the residuals are equal to or less than a threshold value (step S620). Specifically, the separation line determination unit 27 determines whether all the residuals are equal to or less than a threshold value. The threshold value is, for example, 1 meter.

If it is determined that all residuals are equal to or smaller than the threshold (step S620: YES), the separation line determination unit 27 calculates the sum of each residual and the score used to calculate the sum of the residuals (residual sum score) (step S625). If it is determined in the above-mentioned step S620 that at least one residual is greater than the threshold (step S620: NO), the separation line determination unit 27 calculates the sum of each residual, excluding residuals whose residuals are equal to or greater than the threshold, and the score used to calculate the sum of the residuals (residual sum score) (step S630). Specifically, the separation line determination unit 27 calculates the sum of the residuals and the residual sum score of the remaining reverse convex center points after excluding the inverse convex center points (hereinafter also referred to as "outliers") that are far away from each of the lines SL1, SL2, and SL3.

As shown in FIG. 16, in line SL1, point C11 corresponds to an outlier, in line SL2, point C9 corresponds to an outlier, and in line SL3, point C9 corresponds to an outlier. In step S630, these outliers are removed and the sum of the residuals is calculated for each line SL1, SL2, and SL3. Therefore, for example, in the example shown on the left side of FIG. 16, the sum of the residuals for each of points C1, C2, C3, C4, C5, C6, C7, C8, C9, C10, and C12 is calculated. In a general least squares method, the regression line may fluctuate significantly due to the influence of the outliers, so by removing the outliers, the separation line can be determined with high accuracy.

As shown in FIG. 15, after execution of step S625 or after execution of step S630, the separation line determination unit 27 judges whether or not all combinations of points in the inverse convex center point group have been selected as selection points (step S635). If it is judged that all combinations of points have not been selected (step S635: NO), the above-mentioned step S605 is executed. On the other hand, if it is judged that all combinations of points have been selected (step S635: YES), the separation line determination unit 27 determines the line with the smallest residual average value obtained by dividing the sum of residuals by the number of residual sum points as the separation line (step S640). In the example shown in FIG. 16, the residual average value of the line SL3 is smaller than the residual average value of the line SL1 and is also smaller than the residual average value of the line SL2, so the line SL3 is determined as the separation line. After execution of step S640, the separation line determination process ends, and step S70 shown in FIG. 6 is executed.

As shown in FIG. 6, the object recognition unit 26 separates the detection point group (step S70). Specifically, the object recognition unit 26 recognizes the detection point group that is a predetermined distance or more away from the separation line SL as a separate object. The predetermined distance is, for example, 0.3 meters. After step S70 is executed, or if it is determined in the above-mentioned step S50 that the inverse convex shape ITs has not been detected (step S50: NO), the object recognition process ends.

According to the object recognition device 10 of this embodiment having the above configuration, three points selected from the representative point group are the first point P1, the second point P2, and the third point P3, and the inter-vector angle θ between the first vector V1, which is the vector between the first point P1 and the second point P2, and the second vector V2, which is the vector between the second point P2 and the third point P3, is used to determine whether or not the contour point group contains an inverted convex shape ITs. If it is determined that an inverted convex shape ITs is contained, it is recognized that a linear object Ob1 and an object Ob3 that is different from the linear object are present close to each other, and therefore multiple objects that are present close to each other can be accurately distinguished.

The representative point group acquisition unit 24 sequentially sets a plurality of contour points RP included in the contour point group as a reference point SP, and sequentially extracts contour points that are a predetermined distance d1 or more from the reference point SP as representative points to acquire a representative point group, thereby suppressing the occurrence of non-detection and erroneous detection of the inverse convex shape ITs. The inverse convexity determination unit 25 sequentially sets a plurality of representative points included in the representative point group as a first point P1 to calculate an inter-vector angle θ, and determines that an inverse convex shape ITs is included when an inter-vector angle θ that is a predetermined angle or more is detected among the inter-vector angles θ, so that the inverse convex shape ITs can be easily and accurately detected. The separation line determination unit 27 determines a separation line using the inverse convex center point group, i.e., the points that constitute the convex part in the inverse convex shape ITs, so that the separation line can be determined more accurately than a configuration in which a separation line is determined using all points in the representative point group. The separation line determination unit 27 sequentially selects two points from the group of inverse convex center points, calculates the sum of the residuals between the lines SL1, SL2, and SL3 that pass through the two selected points and each of the inverse convex center points, and determines the line with the smallest sum of residuals as the separation line, so that the separation line can be determined more accurately than in a configuration that determines the separation line using a general least squares method.

B. Other embodiments:
(1) In the above embodiment, when there are multiple points that are at a distance d1 or more from the reference point SP, the representative point group acquisition unit 24 may extract the point at which the inter-vector angle θ is maximum as the representative point.

In FIG. 17, the situation where the next representative point is searched for after the representative points Dp41 and Dp42 are extracted is shown. First, the representative point group acquisition unit 24 searches for a point where the inter-vector angle θ is maximum. Specifically, the representative point group acquisition unit 24 searches for a contour point where the angle between the vector V1 connecting the representative point Dp41 and the representative point Dp42 and the vector connecting the representative point Dp42 and the candidate point of the representative point is maximum in the convex direction. For example, in the case of the first candidate point K1, the angle between the vector V1 and the vector V4 connecting the representative point Dp42 and the first candidate point K1 is not large in the reverse convex direction. In contrast, in the case of the second candidate point K2, the angle between the vector V1 and the vector V5 connecting the representative point Dp42 and the second candidate point K2 is large in the reverse convex direction. Therefore, the representative point group acquisition unit 24 extracts the second candidate point K2 as a representative point. In this case, if a contour point closer to the representative point Dp42 is selected as the representative point, the accuracy of the reverse convexity determination may decrease, so the distance between the representative point Dp42 and the candidate points K1 and K2 must be a predetermined distance d1 or more. With this configuration, the representative point can be extracted more accurately and the decrease in accuracy of the reverse convexity determination can be suppressed compared to a configuration in which a point that is a predetermined distance d1 or more from the representative point Dp42 is extracted as the representative point.

(2) In each of the above embodiments, the separation line is determined using only the inverse convex center points of the inverse convex center point group whose residual is smaller than a threshold value. However, the separation line may be determined using all the inverse convex center points of the inverse convex center point group. In addition, the separation line may be determined using the least squares method to reduce calculation costs.

(3) In each of the above embodiments, the object recognition device 10 may omit the separation line determination unit 27. Also, the object recognition device 10 is connected to the light emission control unit 41 and the light reception control unit 51 via the input/output interface 11. That is, the object recognition device 10 and the light emission and reception of the irradiation light Lz are configured as separate entities, but the object recognition device 10 may also include the light emission control unit 41, the light reception control unit 51, the light emitting element 42, and the light receiving element 52.

Each unit of the control unit and the like described in the present disclosure and the method thereof may be realized by a dedicated computer provided by configuring a processor and memory programmed to execute one or more functions embodied in a computer program. Alternatively, each unit of the control unit and the like described in the present disclosure and the method thereof may be realized by a dedicated computer provided by configuring a processor with one or more dedicated hardware logic circuits. Alternatively, the control unit and the method thereof described in the present disclosure may be realized by one or more dedicated computers configured by a combination of a processor and memory programmed to execute one or more functions and a processor configured with one or more hardware logic circuits. In addition, the computer program may be stored in a computer-readable non-transient tangible recording medium as instructions executed by the computer.

The present disclosure is not limited to the above-described embodiments, and can be realized in various configurations without departing from the spirit of the present disclosure. For example, the technical features in the embodiments corresponding to the technical features in each form described in the Summary of the Invention column can be replaced or combined as appropriate to solve some or all of the above-described problems or to achieve some or all of the above-described effects. Furthermore, if a technical feature is not described as essential in this specification, it can be deleted as appropriate.

10... object recognition device, 22... detection point cloud acquisition unit, 23... contour point cloud acquisition unit, 24... representative point cloud acquisition unit, 25... inverse convexity determination unit, 26... object recognition unit, DP... representative point, ITs... inverse convex shape, KP... detection point, Ob1, Ob2, Ob3, Ob4... object, P1... first point, P2... second point, P3... third point, RP... contour point, V1... first vector, V2... second vector, θ... angle between vectors

Claims (7)

An object recognition device (10),
a detection point cloud acquisition unit (22) that acquires a detection point cloud composed of a plurality of detection points (KP) that represent objects (Ob1, Ob2, Ob3, Ob4) existing around the object recognition device;
a contour point group acquisition unit (23) for extracting contour points (RP) of the object from the detection point group to acquire a contour point group;
a representative point group acquisition unit (24) for acquiring a representative point group by extracting representative points (DP) from the contour point group;
an inverse convexity determination unit (25) that determines whether or not an inverse convex shape (ITs) is included in the contour point group by using three points selected from the representative point group as a first point (P1), a second point (P2), and a third point (P3), and a vector angle (θ) between a first vector (V1) that is a vector between the first point and the second point, and a second vector (V2) that is a vector between the second point and the third point;
an object recognition unit (26) that, when it is determined that the inverted convex shape is included, separates the detection point group into a first detection point group (G1) that represents a wall, which is a linear object among the objects present around the object recognition device, and a second detection point group (G2) that represents an object different from the wall and that exists in the vicinity of the wall;
An object recognition device comprising: The object recognition device according to claim 1 ,
The representative point group acquisition unit
a plurality of the contour points included in the contour point group are sequentially set as reference points (SP), and the contour points that are a predetermined distance (d1) or more from the reference point are sequentially extracted as the representative points to obtain the representative point group;
Object recognition device. The object recognition device according to claim 2,
the representative point group acquisition unit extracts, as the representative point, a point that is a distance from the reference point that is equal to or greater than the predetermined distance and at which the inter-vector angle is maximum.
Object recognition device. The object recognition device according to any one of claims 2 to 3,
The reverse convexity determination unit
Sequentially setting a plurality of the representative points included in the representative point group to the first point, and calculating the angle between the vectors;
determining that the inverted convex shape is included when an angle between vectors that is equal to or greater than a predetermined angle is detected among the angles between vectors;
Object recognition device. The object recognition device according to any one of claims 2 to 4,
a separation line determination unit (27) that determines a separation line for separating the detection point group into the first detection point group (G1) and the second detection point group (G2) using an inverse convex center point group constituted by a plurality of the inverse convex center points with each of the second points as an inverse convex center point,
Object recognition device. The object recognition device according to claim 5 ,
The separation line determination unit
Select two points from the group of inverse convex center points in sequence;
Calculating the sum of the residuals between a straight line passing through the two selected points and each of the inverse convex center points;
the straight line having the smallest average residual error obtained by dividing the sum of the residual errors by the number of points used to calculate the sum of the residual errors is determined as the separating straight line;
Object recognition device. An object recognition method executed by an object recognition device, comprising:
A step of acquiring a detection point group composed of a plurality of detection points (KP) representing objects (Ob1, Ob2, Ob3, Ob4) existing around the object recognition device;
extracting contour points (RP) of the object from the detection point cloud to obtain a contour point cloud;
A step of obtaining a representative point group by extracting a representative point (DP) from the contour point group;
a step of determining whether or not an inverted convex shape (ITs) is included in the contour point group by using an inter-vector angle (θ) between a first vector (V1) that is a vector between the first point and the second point, and a second vector (V2) that is a vector between the second point and the third point, and designating three points selected from the representative point group as a first point (P1), a second point (P2), and a third point (P3);
When it is determined that the inverted convex shape is included, separating the detection point group into a first detection point group (G1) representing a wall that is a linear object among the objects present around the object recognition device, and a second detection point group (G2) representing an object different from the wall and present in the vicinity of the wall;
The object recognition method includes:

Images