Shape estimation method, program, shape estimation device and autonomous mobile body

Abstract

To provide a shape estimation method capable of accurately estimating a shape of an object which has a connection part.SOLUTION: A shape estimation method for estimating a shape of an object, includes: an observation value acquisition step S10 of acquiring multiple observation values which indicate a position of the object, by observing the object from an observation point; a clustering step S20 of clustering the multiple observation values for generating a cluster; a detection step S40 of detecting one or more inflection points of the cluster; a division step S50 of dividing the cluster into multiple divided clusters, on one or more inflection points, when one or more inflection points are detected in the detection step; and an estimation step S60 of estimating a shape of a part corresponding to each of the multiple divided cluster, in the object.SELECTED DRAWING: Figure 13

Description

[Technical field]

[0001] The present invention relates to a shape estimation method, a program, a shape estimation device, and an autonomous moving body. [Background technology]

[0002] Currently, research into the automatic operation of moving bodies such as automobiles is being actively conducted. In the automatic operation of a moving body, it is necessary to estimate the shape and position of the other moving body in order to avoid collision with the other moving body. For example, Patent Document 1 describes a technology for clustering a point cloud indicating the other moving body based on a plurality of observation values ​​obtained by observing the other moving body by LiDAR (Light Detection and Ranging). The invention described in Patent Document 1 attempts to estimate the shape of the other moving body based on the clustered point cloud. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] China Patent Application Publication No. 116109601 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the invention described in Patent Document 1, it is not possible to accurately estimate the shape and position of a moving body that has an articulation part and can bend at the articulation part, such as a trailer truck. In the invention described in Patent Document 1, the shape of the moving body in a plan view is approximated by a single rectangular frame regardless of whether the moving body has an articulation part and is bent at the articulation part or not. Therefore, in the invention described in Patent Document 1, for example, when the moving body is bent at the articulation part, it is not possible to accurately estimate the shape of the moving body.

[0005] Therefore, an object of the present invention is to provide a shape estimation method capable of estimating the shape of an object having a joint with high accuracy. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, a shape estimation method according to one embodiment of the present invention is a shape estimation method for estimating the shape of an object, and includes an observation value acquisition process for acquiring a plurality of observation values ​​indicating a position of the object by observing the object from an observation point, a clustering process for creating a cluster by clustering the plurality of observation values, a detection process for detecting one or more bending points of the cluster, a division process for dividing the cluster at the one or more bending points if the one or more bending points are detected in the detection process, and an estimation process for estimating the shape of portions of the object corresponding to each of the plurality of divided clusters.

[0007] In order to achieve the above object, a program according to one aspect of the present invention is a program for causing a computer to execute the above shape estimation method.

[0008] In addition, in order to achieve the above-mentioned object, a shape estimation device according to one embodiment of the present invention is a shape estimation device that estimates the shape of an object, and includes an observation value acquisition unit that acquires a plurality of observation values ​​indicating a position of the object by observing the object from an observation point, a clustering unit that creates a cluster by clustering the plurality of observation values, a detection unit that detects one or more bend points of the cluster, a division unit that, when the one or more bend points are detected by the detection unit, divides the cluster into a plurality of split clusters at the one or more bend points, and an estimation unit that estimates the shape of portions of the object corresponding to each of the plurality of split clusters.

[0009] In order to achieve the above object, an autonomous moving body according to one aspect of the present invention includes the above shape estimation device and an observation value acquisition unit that acquires the multiple observation values.

[0010] In addition, these comprehensive or specific aspects may be realized by a system, a method, an integrated circuit, a computer program, or a non-transitory computer-readable recording medium such as a CD-ROM, or may be realized by any combination of a system, a method, an integrated circuit, a computer program, and a recording medium. Effect of the Invention

[0011] According to the present invention, it is possible to provide a shape estimation method capable of estimating the shape of an object having a joint with high accuracy. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing an example of a configuration of a shape estimation device and an autonomous moving body according to an embodiment; [Diagram 2] FIG. 2 is a diagram showing an example of a first convex hull in the embodiment; [Diagram 3] 4A and 4B are diagrams illustrating an example of a first convex hull and contour points in the embodiment; [Figure 4] FIG. 13 is a diagram showing visible line segments of a first convex hull in the embodiment. [Diagram 5] FIG. 13 is a diagram showing a line segment obtained by rotating a visible line segment by 180 degrees in the embodiment. [Figure 6] FIG. 2 is a diagram showing a convex hull according to an embodiment; [Figure 7] FIG. 3 is a diagram showing a convex hull corresponding to the first convex hull shown in FIG. 2 . [Figure 8] 6A and 6B are diagrams illustrating an example of a rectangular frame created by a detection unit in the embodiment. [Figure 9] 11 is a plan view showing a method for detecting a plurality of reference points by a detection unit according to an embodiment. FIG. [Figure 10] 11A to 11C are plan views showing a method for creating a plurality of straight lines by a detection unit according to an embodiment. [Figure 11] 11 is a plan view showing a method of dividing a cluster by a dividing unit according to an embodiment; FIG. [Figure 12] 13 is a plan view showing a method for estimating the shape of each divided cluster by an estimation unit according to the embodiment; FIG. [Figure 13] 4 is a flowchart showing a flow of a shape estimation method according to the embodiment. [Figure 14] 11A to 11C are diagrams illustrating an example of a shape estimation of an object by a shape estimation method according to a comparative example. [Figure 15] 1A to 1C are diagrams illustrating an example of a shape estimation of an object by a shape estimation method according to an embodiment. [Figure 16] 11 is a graph showing distances from each autonomous moving body to an object when a shape estimation method according to a comparative example and an embodiment is used; [Figure 17] 10 is a graph showing errors in the positions of bending points detected in the shape estimation method according to the embodiment; [Figure 18] 11A to 11C are diagrams showing estimation results obtained when observation values ​​of almost the entire trailer truck, which is an object, are obtained in the shape estimation method according to the embodiment. [Figure 19] 11 is a graph showing Simpson's coefficients for a first estimation result obtained by the shape estimation method according to the embodiment; [Figure 20] 11 is a graph showing IoU for a first estimation result by the shape estimation method according to the embodiment. [Figure 21] 13 is a graph showing the Simpson coefficient for a second estimation result by the shape estimation method according to the embodiment. [Figure 22] 13 is a graph showing IoU for a second estimation result by the shape estimation method according to the embodiment. [Diagram 23] 11 is a schematic plan view showing a first distribution example of a plurality of observed values ​​constituting a divided cluster; FIG. [Figure 24] 13 is a schematic plan view showing a second distribution example of a plurality of observed values ​​constituting a divided cluster; FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

[0014] The embodiments described below are all comprehensive or specific examples. The numerical values, shapes, materials, components, component arrangement and connection forms, steps, and order of steps shown in the following embodiments are merely examples and are not intended to limit the present invention. Furthermore, among the components in the following embodiments, components that are not described in an independent claim showing a top concept are described as optional components.

[0015] In addition, each drawing is a schematic diagram and is not necessarily a precise illustration. In addition, in each drawing, the same components are denoted by the same reference numerals.

[0016] (Embodiment) A shape estimation method, a shape estimation device, and an autonomous moving body according to an embodiment will be described.

[0017] [1. Configuration of shape estimation device and autonomous moving body] The configurations of a shape estimation device and an autonomous moving body according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the configurations of a shape estimation device 100 and an autonomous moving body V10 according to this embodiment.

[0018] The autonomous moving body V10 is a moving body that performs automatic driving, and includes an observation value acquisition unit 10 and a shape estimation device 100. In this embodiment, the autonomous moving body V10 further includes a travel control unit 70 and a memory unit 80. The autonomous moving body V10 is not particularly limited as long as it is a moving body. In this embodiment, the autonomous moving body V10 is a vehicle that performs automatic driving. Hereinafter, the autonomous moving body V10 is also referred to as the host vehicle.

[0019] The shape estimation device 100 is a device that estimates the shape of an object. The shape estimation device 100 according to this embodiment estimates the shape of an object using a shape estimation method described later. The shape estimation device 100 estimates the shape of a moving body around an autonomous moving body V10 as an example of the shape of the object. As shown in FIG. 1, the shape estimation device 100 includes a clustering unit 20, a convex hull creation unit 30, a detection unit 40, a division unit 50, and an estimation unit 60. Note that the shape estimation device 100 may include other processing units, etc. For example, the shape estimation device 100 may include an observation value acquisition unit 10, etc.

[0020] The observation value acquisition unit 10 acquires a plurality of observation values ​​indicating the position of an object by observing the object from an observation point. Here, the observation point is a point on the autonomous moving body V10 where the observation value acquisition unit 10 is arranged. The objects from which the observation value acquisition unit 10 acquires observation values ​​include moving objects and stationary objects around the vehicle. For example, the moving objects include vehicles and pedestrians, and the stationary objects include stopped vehicles, buildings, signs, and the like. In this embodiment, moving objects such as vehicles are mainly handled as objects. The observation value acquired by the observation value acquisition unit 10 includes the direction in which the object is located and the distance to the surface of the object, based on the observation point, which is the installation position of the observation value acquisition unit 10. The observation value acquired by the observation value acquisition unit 10 may also include the height of the object. For example, a three-dimensional information acquisition device such as LiDAR can be used as the observation value acquisition unit 10. In this embodiment, the observation value acquisition unit 10 includes LiDAR.

[0021] The LiDAR measures the distance between the object and the observation point where the LiDAR is installed by irradiating the object with laser light and acquiring the laser light reflected from the object. The observation value acquisition unit 10 acquires three-dimensional point cloud data (point cloud) as height information indicating the relationship between the horizontal position and height of the road surface from the measured distance, the position of the LiDAR, and the direction of the laser light irradiation. The three-dimensional point cloud data is an example of multiple observation values ​​acquired by the observation value acquisition unit 10.

[0022] The LiDAR irradiates, for example, 128 laser beams in the vertical direction (i.e., the vertical direction) at different irradiation angles. The vertical viewing angle of the LiDAR is, for example, -25° or more and 15° or less, with the horizontal direction (i.e., the direction perpendicular to the vertical direction) being 0° (0 degree). The LiDAR can acquire omnidirectional three-dimensional point cloud data by rotating the direction of irradiation of these laser beams by 360° in the horizontal direction. The scan rate of the laser beam is, for example, 5 Hz or more and 20 Hz or less. In this embodiment, the scan rate is set to 10 Hz. That is, in this embodiment, the LiDAR acquires 10 frames of three-dimensional point cloud data per second. The observation value acquisition unit 10 acquires a plurality of observation values ​​corresponding to the target object from the three-dimensional point cloud data.

[0023] The clustering unit 20 is a processing unit that creates clusters by clustering the multiple observation values ​​acquired by the observation value acquisition unit 10. The clustering unit 20 performs a process of combining multiple observation values ​​for each object. For example, the clustering unit 20 combines multiple observation values ​​corresponding to each of multiple moving bodies. This makes it possible to recognize each of the multiple moving bodies. Specifically, for example, when the distance between multiple observation values ​​in the scanning direction of the LiDAR is smaller than a predetermined threshold, the clustering unit 20 combines the multiple observation values ​​to create a block of observation values ​​(also called a run). Next, clusters are created by combining blocks of observation values ​​that are close to each other in a direction perpendicular to the scanning direction of the LiDAR.

[0024] The convex hull creation unit 30 is a processing unit that creates a convex hull that encloses the clusters created by the clustering unit 20. More specifically, the convex hull creation unit 30 creates a convex hull that encloses the clusters in a planar view (i.e., a top view of the object).

[0025] The convex hull created by the convex hull creating unit 30 will be described below with reference to Figs. 2 to 7. Fig. 2 is a diagram showing an example of a first convex hull Ch1 according to this embodiment. Fig. 3 is a diagram showing an example of a first convex hull Ch1 and contour points Ps according to this embodiment. Fig. 4 is a diagram showing a visible line segment Lv of the first convex hull Ch1 according to this embodiment. Fig. 5 is a diagram showing a line segment obtained by rotating the visible line segment Lv by 180° according to this embodiment. Fig. 6 is a diagram showing a convex hull Ch0 according to this embodiment. Fig. 7 is a diagram showing a convex hull Ch0 corresponding to the first convex hull Ch1 shown in Fig. 2.

[0026] As shown in FIG. 2, the convex hull creation unit 30 creates a first convex hull Ch1 surrounding a cluster CL0 created by clustering a plurality of observation values ​​V0 from an observation point P0. The first convex hull Ch1 is a polygon that surrounds the cluster CL0 and is formed so as to have the shortest perimeter. The vertices of the first convex hull Ch1 are also referred to as contour points Ps. As shown in FIG. 2, the shape of the cluster CL0 corresponding to an object such as a vehicle that is actually rectangular in plan view may represent only a part of the object. For example, there are cases where observation values ​​corresponding to a part of the object that is located in the blind spot of the LiDAR cannot be obtained. For this reason, as in the example shown in FIG. 2, the first convex hull Ch1 corresponding to an object that is actually rectangular may be approximately triangular. Therefore, in this embodiment, a method of creating a convex hull having a shape close to the actual shape of the object is adopted. Hereinafter, this method will be described with reference to FIGS. 3 to 6.

[0027] As shown in Fig. 3, the convex hull creation unit 30 creates a first convex hull Ch1 that encloses a cluster CL0 that is created based on a plurality of observation values ​​V0 from an observation point P0 on an autonomous moving body V10, as described with reference to Fig. 2. Note that the plurality of observation values ​​V0 and the cluster CL0 are omitted from illustration in Figs. 3 to 6.

[0028] Next, the convex hull creation unit 30 extracts a visible line segment Lv (a solid line portion of the first convex hull Ch1 shown in FIG. 4) from the first convex hull Ch1, which is a portion facing the observation point P0. In other words, the visible line segment Lv is a portion of the first convex hull Ch1 that is visible from the observation point P0. The first convex hull Ch1 has a visible line segment Lv and an invisible line segment Lr (a dashed line portion of the first convex hull Ch1 shown in FIG. 4). The invisible line segment Lr is a portion of the first convex hull Ch1 that does not face the observation point P0. In other words, the visible line segment Lv is located between the invisible line segment Lr and the observation point P0, and the invisible line segment Lr is blocked by the visible line segment Lv and is therefore not visible from the observation point P0.

[0029] Next, the convex hull creation unit 30 creates a line segment Lv1 by rotating the visible line segment Lv by 180 degrees, as shown in Fig. 5. The convex hull creation unit 30 creates the line segment Lv1 by rotating the visible line segment Lv by 180 degrees around the midpoint CLv of the line segment connecting both ends of the visible line segment Lv.

[0030] 6, the convex hull creation unit 30 creates a convex hull Ch0 by combining the first convex hull Ch1 with a line segment Lv1 obtained by rotating the visible line segment Lv facing the observation point P0 of the first convex hull Ch1 by 180°. In other words, the convex hull creation unit 30 creates a convex hull Ch0 that surrounds the first convex hull Ch1 and the line segment Lv1.

[0031] By creating such a convex hull Ch0, it is possible to create a convex hull Ch0 that is close to the actual shape of the object in a planar view. For example, for the cluster CL0 shown in Fig. 2, the convex hull Ch0 created by the convex hull creation unit 30 has a shape that is close to a rectangle, which is the actual shape of the object in a planar view, as shown in Fig. 7.

[0032] As described above, the convex hull creation unit 30 according to this embodiment can create a convex hull that takes into account occlusion of the object.

[0033] The detection unit 40 is a processing unit that detects one or more bending points of the clusters created by the clustering unit 20. Below, the processing by the detection unit 40 when the object has one bending point will be described, but similar processing is also possible when the object has two or more bending points.

[0034] First, the detection unit 40 creates a rectangular frame that surrounds the convex hull created by the convex hull creation unit 30. This rectangular frame will be described with reference to Fig. 8. Fig. 8 is a diagram showing an example of a rectangular frame Fr0 created by the detection unit 40 according to this embodiment. Note that Fig. 8 shows the convex hull Ch0 shown in Fig. 7 and the rectangular frame Fr0 that surrounds it. As shown in Fig. 8, the detection unit 40 creates the smallest rectangular frame Fr0 that surrounds the convex hull Ch0.

[0035] Next, the detection unit 40 detects a plurality of reference points. The method of detecting the plurality of reference points will be described with reference to Fig. 9. Fig. 9 is a plan view showing the method of detecting the plurality of reference points Pr by the detection unit 40 according to the present embodiment.

[0036] Fig. 9 shows a plan view of a cluster CL0 obtained by clustering a plurality of observation values ​​obtained by observing an object, and a rectangular frame Fr0 surrounding the cluster CL0. Fig. 9 shows a cluster CL0 obtained when a trailer truck having one joint is observed as an example of an object. The rectangular frame Fr0 is created in the same manner as the rectangular frame Fr0 shown in Fig. 8.

[0037] A trailer truck having one articulation is a moving body in which a towing vehicle and a trailer towed by the towing vehicle are articulated at the articulation. The trailer truck bends at the articulation when changing the direction of travel. Figure 9 shows a cluster CL0 created based on a plurality of observations showing a state in which the trailer truck is bent at the articulation.

[0038] The detection unit 40 divides the cluster CL0 with a plurality of division lines Ld that intersect with the longitudinal direction of the cluster CL0. In this embodiment, the detection unit 40 divides the cluster CL0 with a plurality of division lines Ld that are perpendicular to the longitudinal direction of the rectangular frame Fr0. The plurality of division lines Ld may be disposed at equal intervals in the longitudinal direction of the cluster CL0. This allows the position of the side surface of the object to be accurately approximated by a plurality of straight lines. The number of the plurality of division lines Ld is not particularly limited, but may be determined so that the interval between adjacent division lines Ld is 50 cm or more and 100 cm or less. In this embodiment, the number of the plurality of division lines Ld is 20. Note that in FIG. 9 and the like, only some of the division lines Ld and some of the reference points Pr are shown to avoid the drawing becoming complicated.

[0039] Each of the multiple reference points Pr detected by the detection unit 40 is a point that corresponds to an observation value that is located on one of the multiple division lines Ld and is closest to the observation point P0 among the multiple observation values. The multiple reference points Pr correspond to points on the side of the contour of the cluster CL0 that are close to the observation point P0 of the object. Note that the reference point Pr being located on the division line Ld does not have to be interpreted as a state in which the position of the reference point Pr and the position of the division line Ld completely match. For example, a state in which the distance of the reference point Pr from the division line Ld is less than a predetermined threshold value may be expressed as the reference point being located on the division line. For example, the predetermined threshold value may be about the accuracy of the position of the observation value by the LiDAR (i.e., the position resolution).

[0040] Next, the detection unit 40 approximates a portion of the contour of the cluster CL0 that corresponds to a side surface close to the observation point P0 of the object with a plurality of straight lines. Each of the one or more bending points is detected based on an intersection of two straight lines among the plurality of straight lines. Hereinafter, a method of creating such a plurality of straight lines will be described with reference to FIG. 10. FIG. 10 is a plan view showing a method of creating a plurality of straight lines by the detection unit 40 according to the present embodiment. In this embodiment, the plurality of straight lines are determined based on a plurality of reference points Pr on a plurality of division lines Ld. For example, in the example shown in FIG. 10, a straight line L1 is created as an approximation straight line based on a reference point Pr (reference point Pr located at the left end of FIG. 10) located at one end of the plurality of reference points Pr and two or more reference points Pr including the reference point Pr and the reference point Pr adjacent to the reference point Pr. In addition, a straight line L2 is created as an approximation straight line based on a reference point Pr (reference point Pr located at the right end of FIG. 10) located at the other end of the plurality of reference points Pr and two or more reference points Pr including the reference point Pr and the reference point Pr adjacent to the reference point Pr. In this way, by determining multiple straight lines based on multiple reference points Pr, the position of the side of the object closer to the observation point P0 can be accurately approximated by the multiple straight lines in either case where the side of the object closer to the observation point P0 is convexly curved or concavely curved.

[0041] The approximation line is created using, for example, the least squares method. In this embodiment, a bending point is detected based on the intersection Pc of the two straight lines L1 and L2. In this way, when the object has a junction, the bending point corresponding to the position of the junction can be detected with high accuracy.

[0042] In the detection unit 40, each of the one or more bending points may be detected when the intersection angle θc of the two straight lines is within a predetermined range. For example, each of the one or more bending points may be detected based on the intersection point of the two straight lines when the intersection angle of the two straight lines is greater than 0.6π and less than 0.98π, and may not be detected based on the intersection point of the two straight lines when the intersection angle of the two straight lines is 0.6π or less or 0.98π or more. By not detecting the bending point when the intersection angle is 0.6π or less, it is possible to reduce erroneous detection of the bending point when the target object is a vehicle or the like that does not have a joint. In addition, by not detecting the bending point when the intersection angle is 0.98π or more, it is possible to reduce detection of the bending point in a vehicle that moves straight (not bent). In addition, when the intersection angle is close to π (180°), the error in the position of the bending point is likely to be large, so by not detecting the bending point when the intersection angle is 0.98π or more, it is possible to reduce the error in the position of the bending point.

[0043] Furthermore, each of the one or more bend points may be detected based on the intersection point of the two lines when the intersection angle of the two lines is greater than 0.65π and less than 0.95π, and may not be detected based on the intersection point of the two lines when the intersection angle of the two lines is 0.65π or less or 0.95π or more, thereby making it possible to further reduce erroneous detection of bend points and further reduce errors in the positions of the bend points.

[0044] The detection unit 40 may not detect the bending point when the size of the cluster CL0 is equal to or smaller than a predetermined threshold. This can reduce the erroneous detection of the bending point when the target object is a moving object without a joint, such as a small car. The detection unit 40 may not detect the bending point when the length of the cluster is, for example, 10 m or less.

[0045] The intersection point Pc of the two straight lines detected as described above may be itself the bending point, but the bending point may also be detected using a low-pass filter based on information on bending points detected in the past. For example, at a certain detection timing t nThe intersection point of the two lines obtained from multiple observations at y(t n ) and the detection timing t n The detection timing t immediately before n-1 The position of the inflection point detected in x(t n-1 ), the detection timing t n Position of the bending point x(t n ) may be expressed by the following formula (1):

[0046] x(t n )=x(t n-1 )+{y(t n )-x(t n-1 )}×ka (1)

[0047] Here, ka is the position of the bending point x(t n ) position, the intersection point y(t n ), and the position of the previously detected bending point x(t n-1 ) is a constant corresponding to a weighting factor by which the weighting factor k is multiplied. The constant k is set to, for example, a value equal to or greater than 0.9 and less than 1. In this embodiment, the constant k is set to 0.9. In this manner, the bending point may be updated using information on past bending points. In other words, when one or more bending points are detected by the detection unit 40, one or more bending points of the object at the subsequent detection timing may be detected using information on the detected one or more bending points. This makes it possible to suppress an increase in the error in the position of the bending point due to noise or the like.

[0048] The dividing unit 50 is a processing unit that divides the cluster CL0 into a plurality of divided clusters at one or more bending points when one or more bending points are detected by the detection unit 40. The processing by the dividing unit 50 will be described with reference to Fig. 11. Fig. 11 is a plan view showing a method of dividing the cluster CL0 by the dividing unit 50 according to the present embodiment.

[0049] The dividing unit 50 divides the cluster CL0 at the bending point Pb shown in Fig. 11. In the example shown in Fig. 11, the dividing unit 50 divides the cluster CL0 into a divided cluster CL1 and a divided cluster CL2 by a straight line Ldc passing through the bending point Pb. In this embodiment, the straight line Ldc passes through the bending point Pb and is parallel to the bisector of the angle formed by the two straight lines L1 and L2. Observations located on the straight line Ldc including the bending point Pb may be included in each divided cluster.

[0050] The estimation unit 60 is a processing unit that estimates the shape of the object. When one or more bending points of the cluster CL0 are not detected, the estimation unit 60 estimates one shape indicating the entire object. In this embodiment, when one or more bending points of the cluster CL0 are not detected, the estimation unit 60 estimates the shape of the object to be a rectangular frame Fr0 as shown in Figs. 8 and 9.

[0051] When one or more bending points of cluster CL0 are detected, estimation unit 60 estimates the shape of the portion of the object corresponding to each of the divided clusters. The process performed by estimation unit 60 according to this embodiment in such a case will be described below with reference to Fig. 12. Fig. 12 is a plan view showing a method of estimating the shape of each divided cluster by estimation unit 60 according to this embodiment.

[0052] The estimation unit 60 creates rectangular frames as estimated shapes of portions of the object corresponding to each divided cluster. Specifically, the estimation unit 60 creates a convex hull surrounding each divided cluster in the same manner as the convex hull creation unit 30, and creates the smallest rectangular frame surrounding the convex hull in the same manner as the detection unit 40. In this way, the estimation unit 60 estimates the shapes of the divided clusters CL1 and CL2 in a planar view to be rectangular frames Fr1 and Fr2, respectively.

[0053] The driving control unit 70 is a processing unit that controls the driving of the host vehicle. In this embodiment, the driving control unit 70 controls the driving so as to avoid collision with surrounding vehicles based on the shapes of surrounding vehicles and the like estimated by the shape estimation device 100. The driving control unit 70 controls the traveling direction of the host vehicle and the like based on, for example, the position of the host vehicle, map information stored in the storage unit 80, and the like.

[0054] The storage unit 80 stores information for estimating the position of the vehicle. For example, the storage unit 80 stores map information including information on the road on which the vehicle is driven. The storage unit 80 may also store information for automatic driving.

[0055] [2. Shape estimation method] The shape estimation method of this embodiment will be described with reference to Fig. 13. Fig. 13 is a flowchart showing the flow of the shape estimation method according to this embodiment.

[0056] The shape estimation method according to this embodiment is a method for estimating the shape of an object such as a moving body. In the shape estimation method according to this embodiment, first, as shown in Fig. 13, the observation value acquisition unit 10 of the shape estimation device 100 acquires a plurality of observation values ​​indicating the position of the object by observing the object from an observation point (observation value acquisition step S10). In this embodiment, the observation value acquisition unit 10 acquires three-dimensional point cloud data as observation values ​​using a LiDAR.

[0057] Next, the clustering unit 20 creates clusters by clustering the multiple observed values ​​(clustering step S20).

[0058] Next, the convex hull creation unit 30 creates a convex hull surrounding the cluster (convex hull creation step S30). In this embodiment, in the convex hull creation step S30, a first convex hull surrounding the cluster is created, and the convex hull is created by combining the first convex hull with a line segment of the first convex hull that is obtained by rotating a visible line segment facing the observation point by 180 degrees. This makes it possible to create a convex hull Ch0 that is close to the actual shape of the object in a planar view.

[0059] Next, the detection unit 40 detects one or more bending points of the cluster (detection step S40). Specifically, in the detection step S40, the detection unit 40 approximates a portion of the outline of the cluster that corresponds to a side surface close to the observation point of the object with a plurality of straight lines, and each of the one or more bending points is detected based on an intersection of two straight lines among the plurality of straight lines. This makes it possible to accurately detect bending points corresponding to the positions of the connecting parts when the object has a connecting part like a trailer truck.

[0060] In this embodiment, in the detection step S40, the detection unit 40 divides the cluster by a plurality of division lines intersecting the longitudinal direction of the cluster. A plurality of straight lines approximating a portion of the outline of the cluster that corresponds to the side surface of the object close to the observation point are determined based on a plurality of reference points on the plurality of division lines. Each of the plurality of reference points corresponds to an observation value that is located on one of the division lines included in the plurality of division lines and is closest to the observation point among the plurality of observation values. By determining the plurality of straight lines based on such a plurality of reference points, the position of the side surface of the object can be accurately approximated by the plurality of straight lines in both cases where the side surface of the object close to the observation point is bent convexly and where the side surface is bent concavely.

[0061] In the detection step S40, the multiple division lines may be arranged at equal intervals in the longitudinal direction of the cluster, so that the positions of the side faces of the object can be accurately approximated by the multiple straight lines.

[0062] Furthermore, in the detection step S40, as described above, each of the one or more bending points may be detected using a low-pass filter. In this way, when one or more bending points are detected in the detection step S40, one or more bending points of the object at the subsequent detection timings may be detected using information on the detected one or more bending points. This makes it possible to prevent the position error of the bending points from becoming large due to noise, etc.

[0063] In the detection step S40, when the intersection angle of two straight lines approximating a portion corresponding to a side surface close to the observation point of the object is greater than 0.6π and less than 0.98π, each of the one or more bending points is detected based on the intersection point, and when the intersection angle of the two straight lines is 0.6π or less, or 0.98π or more, each of the one or more bending points is not detected based on the intersection point. By not detecting the bending point when the intersection angle is 0.6π or less, it is possible to reduce erroneous detection of the bending point when the object is a vehicle or the like that does not have a joint. In addition, by not detecting the bending point when the intersection angle is 0.98π or more, it is possible to reduce detection of the bending point in a vehicle that moves straight (not bent). In addition, when the intersection angle is close to π (180°), the error in the position of the bending point is likely to be large, so by not detecting the bending point when the intersection angle is 0.98π or more, it is possible to reduce the error in the position of the bending point.

[0064] Furthermore, each of the one or more bend points may be detected based on the intersection point of the two lines when the intersection angle of the two lines is greater than 0.65π and less than 0.95π, and may not be detected based on the intersection point of the two lines when the intersection angle of the two lines is 0.65π or less or 0.95π or more, thereby making it possible to further reduce erroneous detection of bend points and further reduce errors in the positions of the bend points.

[0065] In the detection step S40, if the size of the cluster is equal to or smaller than a predetermined threshold, it is not necessary to detect the bending point. This can reduce erroneous detection of the bending point when the target object is a moving object without a joint, such as a small car. In the detection step S40, if the length of the cluster is, for example, 10 m or less, it is not necessary to detect the bending point.

[0066] If one or more bending points are detected in the detection step S40 (Yes in the detection step S40), the division unit 50 divides the cluster into a plurality of divided clusters at the one or more bending points (division step S50). In the division step S50, for example, the cluster may be divided by a straight line passing through the bending point. The straight line may be a straight line parallel to the bisector of the angle formed by the two straight lines used to detect the bending points.

[0067] Next, the estimation unit 60 estimates the shape of the object (estimation step S60). In the division step S50, when the cluster is divided into a plurality of divided clusters, the estimation unit 60 estimates the shape of a portion of the object corresponding to each of the plurality of divided clusters. The estimation unit 60 creates a convex hull surrounding each divided cluster in the same manner as the convex hull creation unit 30, and creates the smallest rectangular frame surrounding the convex hull in the same manner as the detection unit 40. In this way, the estimation unit 60 estimates the shape of each divided cluster in a planar view to be a rectangular frame.

[0068] If one or more bending points are not detected in the detection step S40 (No in the detection step S40), the process proceeds to the estimation step S60, where the estimation unit 60 estimates the shape of the object. In this embodiment, the estimation unit 60 creates a rectangular frame surrounding the cluster as the estimated shape of the object. More specifically, the estimation unit 60 estimates that the rectangular frame surrounding the convex hull created by the convex hull creation unit 30 is the shape of the object.

[0069] Then, the process returns to the observation value acquisition step S10, and the same steps are repeated.

[0070] By using the above-described shape estimation method, the shape of a moving object having a joint can be estimated with high accuracy.

[0071] [3. Effects] The effects of the shape estimation method (and shape estimation device 100) according to this embodiment will be described with reference to experimental results.

[0072] The estimation results of the distance from an autonomous moving body V10 to an object when the shape estimation method of this embodiment is used will be explained using Figures 14 to 16, while comparing them with the estimation results of the distance from an autonomous moving body V12 to an object when a comparative example is used.

[0073] FIG. 14 and FIG. 15 are diagrams showing examples of shape estimation of an object by a comparative example and a shape estimation method according to the present embodiment, respectively. FIG. 14 and FIG. 15 show an experimental example in which the shape of a trailer truck running parallel to each autonomous moving body without colliding with the autonomous moving body is estimated as an object. In addition, FIG. 14 and FIG. 15 also show autonomous moving bodies V12 and V10 at which observation points are located, respectively. FIG. 16 is a graph showing the distance from each autonomous moving body to an object when a shape estimation method according to the comparative example and the present embodiment is used. Here, the distance from each autonomous moving body to an object means the shortest distance from the side of each autonomous moving body to the edge of the object. FIG. 16 shows the distance from each autonomous moving body to an object when each autonomous moving body and the trailer truck, which is the object, turn right while running parallel to each other. The horizontal axis of FIG. 16 shows time. In addition, in Figure 16, the distance from the autonomous moving body V10 to the target object, which is the towing vehicle of the trailer truck, is shown by a solid line, and the distance from the autonomous moving body V10 to the trailer (towed vehicle) of the trailer truck is shown by a dashed line, as the distance from the autonomous moving body V10 to the target object, which is the towing vehicle of the trailer truck.

[0074] The shape estimation method of the comparative example is a method in which a rectangular frame surrounding the convex hull created in the convex hull creation step S30 according to the present embodiment is taken as the estimated shape of the object. That is, in the shape estimation method of the comparative example, the shape of the object is estimated as a single rectangular frame.

[0075] As shown in FIG. 14, in the shape estimation method of the comparative example, the shape of the trailer truck, which is the object, in a plan view is estimated as a rectangular frame Fr0 surrounding a cluster CL0 created based on a plurality of observation values. Therefore, although there is actually a gap between the autonomous moving body V12 and the cluster CL0, the rectangular frame Fr0 and the autonomous moving body V12 overlap. Therefore, as shown in FIG. 16, in the shape estimation method of the comparative example, it is erroneously estimated that the object and the autonomous moving body V12 are colliding (the distance is 0 or less). When such a shape estimation method of the comparative example is used, the positional relationship between the autonomous moving body V12 and the moving body, which is the object located in the vicinity, cannot be correctly estimated, and therefore appropriate automatic driving cannot be performed.

[0076] On the other hand, as shown in FIG. 15, in the shape estimation method according to the present embodiment, the cluster is divided at the bending point, so that the shape of the object having the joint can be accurately estimated. More specifically, in the shape estimation method according to the present embodiment, the shape of the object can be accurately estimated by the rectangular frame Fr1 surrounding the divided cluster CL1 and the rectangular frame Fr2 surrounding the divided cluster CL2. Therefore, as shown in FIG. 16, when the shape estimation method according to the present embodiment is used, the distance between the autonomous moving body V10 and the rectangular frames Fr1 and Fr2 is estimated to be 1.5 m or more, in accordance with the positional relationship between the actual autonomous moving body V10 and the towing vehicle and trailer of the object. By using such a shape estimation method according to the present embodiment, the positional relationship between the autonomous moving body V10 and the moving body that is the object located in the periphery can be accurately estimated, so that erroneous determination of a collision between the autonomous moving body V10 and another moving body can be suppressed. Therefore, it becomes possible to perform appropriate automatic driving.

[0077] Here, the accuracy of the position of the bending point detected by the shape estimation method according to the present embodiment will be described with reference to Fig. 17. Fig. 17 is a graph showing the error of the position of the bending point detected by the shape estimation method according to the present embodiment. The horizontal axis of Fig. 17 shows the number of updates of the observation value (i.e., the number of frames updated in the LiDAR).

[0078] As shown in Fig. 17, the error in the position of the bending point is suppressed to 0.65 m or less. As described above with reference to Fig. 16 and other figures, since the distance from the autonomous mobile body V10 to the target object can be estimated with high accuracy, it is considered that the error in the position of this bending point is suppressed to a value small enough to estimate the position of the target object.

[0079] The reason why the errors for the first and sixth observation values ​​in Figure 17 are larger than the errors for the other observation values ​​is presumably because, when these observation values ​​were obtained, the angle of bending at the articulation of the target trailer truck was small (i.e., the intersection angle θc described above was small).

[0080] Next, the evaluation results of the position and size of the rectangular frame, which is the shape estimated by the shape estimation method according to this embodiment, will be described while comparing with the evaluation results of a comparative example. In the shape estimation devices of this embodiment and the comparative example, since the overall height of the trailer truck is higher than the position where the LiDAR is installed, it may not be possible to obtain observation values ​​(three-dimensional point clouds) corresponding to positions other than the side of the object. As a result, as shown in FIG. 15, the size of the rectangular frame estimated as the shape of the object may be smaller than the actual size of the object. Therefore, the area S of the rectangular frame, which is the estimated result of the object, is calculated. Tracking The area S of the union of the estimated rectangular frame and the area where the actual object is located Union , and the area S of the intersection between the estimated rectangular frame and the area in which the actual object is located Intersection The evaluation is performed using IoU (Intersection over Union) shown in the following formula (2) and Simpson's coefficient Cs shown in the following formula (3).

[0081] IoU=S Intersection / S Union (2) Cs=S Intersection / S Tracking (3)

[0082] The IoU and Simpson's coefficient Cs are values ​​between 0 and 1, with a larger value indicating a more appropriate estimation result.

[0083] The following describes the evaluation results of the Simpson's coefficient Cs and IoU of the shape of the object estimated by the shape estimation method according to this embodiment. First, the estimation results to be evaluated will be explained with reference to Fig. 18. Fig. 18 is a diagram showing the estimation results when observation values ​​of almost the entire trailer truck, which is the object, are obtained by the shape estimation method according to this embodiment.

[0084] As shown in FIG. 18, the evaluation results of the Simpson coefficient Cs and IoU in the estimation result (hereinafter also referred to as the first estimation result) in the case where the observation values ​​of the towing vehicle of the trailer truck (corresponding to the divided cluster CL1 and rectangular frame Fr1 shown in FIG. 18) and the trailer (corresponding to the divided cluster CL2 and rectangular frame Fr2) of the target object can be obtained are described with reference to FIG. 19 and FIG. 20. In addition, the evaluation results of the Simpson coefficient Cs and IoU in the estimation result (hereinafter also referred to as the second estimation result) in the case where the observation values ​​of only the side of the trailer truck of the target object can be obtained (see FIG. 15) are described with reference to FIG. 21 and FIG. 22. FIG. 19 and FIG. 20 are graphs showing the Simpson coefficient Cs and IoU for the first estimation result by the shape estimation method according to the present embodiment, respectively. FIG. 21 and FIG. 22 are graphs showing the Simpson coefficient Cs and IoU for the second estimation result by the shape estimation method according to the present embodiment, respectively. The horizontal axis in Fig. 19 to Fig. 22 indicates the number of frames (number of elapsed frames) acquired by LiDAR. Fig. 19 to Fig. 22 also show Simpson's coefficient Cs and IoU for each estimated shape of the towing vehicle and trailer of the trailer truck using the shape estimation method according to the present embodiment. Fig. 19 to Fig. 22 also show Simpson's coefficient Cs and IoU for the estimated shape of the target object using the shape estimation method of the comparative example.

[0085] 19 and 21, the shape estimation result according to this embodiment has an improved Simpson coefficient Cs than the shape estimation result according to the comparative example. Therefore, it can be said that the position of the object can be estimated more accurately in this embodiment than in the comparative example.

[0086] Furthermore, as for IoU, when observation values ​​of the entire object cannot be obtained, as shown in Fig. 22, it is lower than the comparative example, but when observation values ​​of the entire object can be obtained, it is improved compared to the comparative example, as shown in Fig. 20. Therefore, it can be said that the shape estimation method according to this embodiment can estimate size with high accuracy if sufficient observation values ​​are obtained.

[0087] (Variations, etc.) Although the shape estimation method according to one aspect of the present invention has been described based on the embodiment, the present invention is not limited to the embodiment. As long as the embodiment is modified in various ways that are conceivable by a person skilled in the art, the modifications may be included within the scope of the present invention.

[0088] For example, in the above embodiment, the shape of a trailer truck having one joint is estimated as the target object, but the target object is not limited to this. For example, the target object may be a trailer truck, a bus, a streetcar, etc. having multiple joints.

[0089] In the above embodiment, in the convex hull creation step S30, the convex hull creation unit 30 creates a convex hull by combining the first convex hull with a line segment of the first convex hull that is obtained by rotating a visible line segment facing the observation point by 180 degrees, but the configuration of the convex hull creation step S30 is not limited to this. For example, in the convex hull creation step S30, the convex hull creation unit 30 may create the first convex hull itself as a convex hull, without combining a line segment with the first convex hull.

[0090] In the shape estimation device 100 and the shape estimation method according to the above-described embodiment, a convex hull surrounding the cluster is created, but it is not necessary to create a convex hull. For example, a rectangular frame surrounding the cluster may be created without creating a convex hull.

[0091] In addition, in the above embodiment, when the width of the part of the object corresponding to the divided cluster is estimated to be too small, the estimated shape of the object may be corrected. For example, when the object is a vehicle traveling on a road, the width of the object may be corrected when the width is too small compared to the length of the object. Hereinafter, an example of a method for correcting the width of the object will be described with reference to Figs. 23 and 24. Figs. 23 and 24 are schematic plan views showing a first distribution example and a second distribution example of a plurality of observation values ​​constituting a divided cluster, respectively. In Figs. 23 and 24, the actual contour of the object is shown in a dashed frame. In the example shown in Fig. 23, only a plurality of observation values ​​corresponding to one side surface of the object and its vicinity are obtained. In the example shown in Fig. 24, a plurality of observation values ​​corresponding to two side surfaces of the object are obtained.

[0092] First, the width of the divided cluster is calculated. When multiple observation values ​​are obtained as shown in FIG. 23 and FIG. 24, first, the two observation values ​​that are the furthest apart from the multiple observation values ​​are extracted. For example, in the example shown in FIG. 23 and FIG. 24, the observation value P1 and the observation value P2 are extracted. Next, the observation value that is the furthest apart from the straight line that passes through those two observation values ​​is extracted. In the example shown in FIG. 23 and FIG. 24, the observation value P3 is extracted. Next, the distance Lw from the straight line to the observation value P3 is calculated. This distance Lw may be defined as the width of the divided cluster. If the distance Lw is smaller than a predetermined threshold, the width of the object may be corrected. For example, since the width of a light vehicle is 1.48 m or less, if the width of the estimated shape of the object is smaller than 1.48 m, it is presumed that the width of the object is underestimated. For example, if the distance Lw is less than 0.75 m, the width of the estimated shape of the object may be corrected. The estimation unit 60 may search for an observation value whose distance from a line determined based on a reference point corresponding to a divided cluster among the above-mentioned reference points is greater than the width of the divided cluster from among the multiple observation values, and add the searched observation value to the divided cluster. By creating a convex hull and a rectangular frame that surround the divided cluster to which the observation value has been added, the width of the part of the object that corresponds to the divided cluster can be corrected to an appropriate value.

[0093] The following forms may also be included within the scope of one or more aspects of the present disclosure.

[0094] (1) Some of the components included in the above-mentioned shape estimation device 100 may be a computer system including a microprocessor, a ROM, a RAM, a hard disk unit, a display unit, a keyboard, a mouse, etc. A computer program is stored in the RAM or the hard disk unit. The microprocessor operates according to the computer program to achieve its functions. Here, the computer program is composed of a combination of multiple instruction codes that indicate commands for a computer to achieve a predetermined function.

[0095] (2) Some of the components included in the above-described shape estimation device 100 may be configured from one system LSI (Large Scale Integration). The system LSI is an ultra-multifunctional LSI manufactured by integrating multiple components on one chip, and specifically, is a computer system including a microprocessor, ROM, RAM, etc. A computer program is stored in the RAM. The system LSI achieves its functions by the microprocessor operating in accordance with the computer program.

[0096] (3) Some of the components included in the shape estimation device 100 may be configured as an IC card or a standalone module that can be attached to and detached from each device. The IC card or the module is a computer system configured from a microprocessor, ROM, RAM, etc. The IC card or the module may include the ultra-multifunction LSI. The microprocessor operates according to a computer program, causing the IC card or the module to achieve its functions. The IC card or the module may be tamper-resistant.

[0097] (4) Furthermore, some of the components included in the above-described shape estimation device 100 may be the computer program or the digital signal recorded on a computer-readable recording medium, such as a flexible disk, a hard disk, a CD-ROM, an MO, a DVD, a DVD-ROM, a DVD-RAM, a BD (Blu-ray (registered trademark) Disc), a semiconductor memory, etc. Also, they may be the digital signal recorded on such a recording medium.

[0098] In addition, some of the components included in the above-mentioned shape estimation device 100 may transmit the computer program or the digital signal via a telecommunications line, a wireless or wired communication line, a network such as the Internet, data broadcasting, etc.

[0099] (5) The present disclosure may be the shape estimation method described above. It may also be a computer program for causing a computer to realize the shape estimation method described above, or a digital signal consisting of the computer program. Furthermore, the present disclosure may be realized as a non-transitory computer-readable recording medium, such as a CD-ROM, on which the computer program is recorded.

[0100] (6) The present disclosure may also provide a computer system having a microprocessor and a memory, the memory storing the computer program, and the microprocessor operating in accordance with the computer program.

[0101] (7) The program or the digital signal may also be implemented by another independent computer system by recording it on the recording medium and transferring it, or by transferring the program or the digital signal via the network, etc.

[0102] (8) The above-described embodiments and modifications may be combined with each other.

[0103] (Additional Note) Furthermore, based on the above description, the following techniques are disclosed.

[0104] (Technology 1) A shape estimation method for estimating a shape of an object, comprising: an observation value acquisition step of acquiring a plurality of observation values ​​indicating a position of the object by observing the object from an observation point; a clustering step of creating clusters by clustering the plurality of observation values; a detection step of detecting one or more bending points of the clusters; a division step of dividing the cluster at the one or more bending points into a plurality of split clusters if the one or more bending points are detected in the detection step; and an estimation step of estimating shapes of portions of the object corresponding to each of the plurality of split clusters.

[0105] (Technology 2) A shape estimation method according to Technology 1, wherein in the detection step, a portion of the contour of the cluster corresponding to a side of the object close to the observation point is approximated by a plurality of straight lines, and each of the one or more bending points is detected based on an intersection of two straight lines among the plurality of straight lines.

[0106] (Technology 3) A shape estimation method according to Technology 2, wherein in the detection step, when the intersection angle between the two straight lines is greater than 0.65π and less than 0.95π, each of the one or more bend points is detected based on the intersection point, and when the intersection angle between the two straight lines is less than or equal to 0.65π or greater than or equal to 0.95π, each of the one or more bend points is not detected based on the intersection point.

[0107] (Technology 4) A shape estimation method according to Technology 2 or 3, in which, in the detection step, the cluster is divided by a plurality of division lines intersecting the longitudinal direction of the cluster, the plurality of straight lines are determined based on a plurality of reference points on the plurality of division lines, and each of the plurality of reference points corresponds to an observation value among the plurality of observation values ​​that is located on one of the division lines and is closest to the observation point.

[0108] (Technology 5) The shape estimation method according to Technology 4, wherein in the detection step, the multiple division lines are arranged at equal intervals in the longitudinal direction of the cluster.

[0109] (Technology 6) A shape estimation method according to any one of Technologies 1 to 5, further comprising a convex hull creation step of creating a convex hull surrounding the cluster, wherein in the convex hull creation step, a first convex hull surrounding the cluster is created, and the convex hull is created by combining the first convex hull with a line segment of the first convex hull that is obtained by rotating a visible line segment facing the observation point by 180 degrees.

[0110] (Technology 7) The shape estimation method according to any one of Techniques 1 to 6, wherein in the detection step, each of the one or more bending points is detected using a low-pass filter.

[0111] (Technology 8) A shape estimation method according to any one of Techniques 1 to 7, in which, when the one or more bend points of the object are detected in the detection process, information of the detected one or more bend points is used to detect the one or more bend points of the object in the subsequent detection processes.

[0112] (Technology 9) A shape estimation method according to Technology 4 or 5, in which, in the estimation step, when the width of one of the multiple divided clusters is smaller than a predetermined threshold, an observation value whose distance from a straight line determined based on a reference point among the multiple reference points corresponding to the one divided cluster is greater than the width of the one divided cluster is searched from the multiple observation values, and the searched observation value is added to the one divided cluster.

[0113] (Technical Technique 10) A program for causing a computer to execute the shape estimation method according to any one of Technical Techniques 1 to 9.

[0114] Technology (11) A shape estimation device that estimates a shape of an object, the shape estimation device including: a clustering unit that creates clusters by clustering a plurality of observation values ​​that are obtained by observing the object from an observation point, the plurality of observation values ​​indicating a position of the object; a detection unit that detects one or more bending points of the clusters; a division unit that, when the one or more bending points are detected by the detection unit, divides the cluster into a plurality of split clusters at the one or more bending points; and an estimation unit that estimates shapes of portions of the object that correspond to each of the plurality of split clusters.

[0115] Technique (12) An autonomous moving body comprising the shape estimation device according to technique 11 and an observation value acquisition unit that acquires the multiple observation values. [Industrial Applicability]

[0116] A shape estimation method according to an aspect of the present invention can be applied to, for example, vehicles for autonomous driving. [Explanation of symbols]

[0117] 10 Observation value acquisition section 20 Clustering Department 30 Convex hull creation unit 40 Detection unit 50 divisions 60 Estimation part 70 Driving control unit 80 Storage section 100 Shape estimation device Ch0 Convex hull Ch1 First convex hull CL0 Cluster CL1, CL2 split cluster CLv midpoint Fr0, Fr1, Fr2 Rectangular frame Ld Partition line Lv visible line segment Lv1 Line Lr invisible line of sight P0 observation point Pb bending point Pc intersection Pr reference point Ps contour point V0 Observed value V10, V12 Autonomous mobile object

Claims

1. A shape estimation method for estimating a shape of an object, comprising: an observation value acquiring step of acquiring a plurality of observation values ​​indicating a position of the object by observing the object from an observation point; a clustering step of creating clusters by clustering the plurality of observed values; a detection step of detecting one or more inflection points of said cluster; a dividing step of dividing the cluster into a plurality of divided clusters at the one or more bending points when the one or more bending points are detected in the detecting step; and estimating the shape of a portion of the object corresponding to each of the plurality of divided clusters. Shape estimation method.

2. In the detection step, approximating a portion of the outline of the cluster that corresponds to a side surface of the object that is close to the observation point with a plurality of straight lines; Each of the one or more inflection points is detected based on an intersection point of two straight lines among the plurality of straight lines. The shape estimation method according to claim 1 .

3. In the detection step, When an angle between the two straight lines is greater than 0.65π and less than 0.95π, each of the one or more inflection points is detected based on the intersection point; When the intersection angle between the two straight lines is equal to or less than 0.65π or equal to or greater than 0.95π, each of the one or more bending points is not detected based on the intersection point. The shape estimation method according to claim 2 .

4. In the detection step, the cluster is divided by a plurality of division lines intersecting a longitudinal direction of the cluster; the straight lines are determined based on a plurality of reference points on the plurality of dividing lines; Each of the plurality of reference points corresponds to an observation value among the plurality of observation values ​​that is located on one of the division lines and is closest to the observation point. The shape estimation method according to claim 2 or 3.

5. In the detection step, the plurality of division lines are arranged at equal intervals in the longitudinal direction of the cluster. The shape estimation method according to claim 4 .

6. The method further includes a convex hull creation step of creating a convex hull that encloses the cluster; In the convex hull creation step, constructing a first convex hull that encloses the cluster; The convex hull is created by combining the first convex hull with a line segment obtained by rotating a visible line segment facing the observation point by 180 degrees in the first convex hull. The shape estimation method according to claim 1 .

7. In the detecting step, each of the one or more inflection points is detected using a low pass filter. The shape estimation method according to claim 1 .

8. When the one or more bending points of the object are detected in the detection step, the one or more bending points of the object are detected in the subsequent detection steps using information of the one or more detected bending points. The shape estimation method according to claim 1 .

9. In the estimation step, when a width of one of the plurality of divided clusters is smaller than a predetermined threshold, an observation value whose distance from a line determined based on a reference point corresponding to the one divided cluster among the plurality of reference points is greater than the width of the one divided cluster is searched from the plurality of observation values, and the searched observation value is added to the one divided cluster. The shape estimation method according to claim 4 .

10. A method for causing a computer to execute the shape estimation method according to any one of claims 1 to 3 and 6 to 8. program.

11. A shape estimation device that estimates a shape of an object, comprising: an observation value acquisition unit that acquires a plurality of observation values ​​indicating a position of the object by observing the object from an observation point; a clustering unit that creates clusters by clustering the plurality of observed values; a detector for detecting one or more inflection points of the cluster; a division unit that divides the cluster into a plurality of divided clusters at the one or more bending points when the one or more bending points are detected by the detection unit; an estimation unit that estimates a shape of a portion of the object that corresponds to each of the plurality of divided clusters. Shape estimation device.

12. A shape estimation device according to claim 11, an observation value acquisition unit for acquiring the plurality of observation values; Autonomous mobile body.

Images