Object detection method and object detection device
By rearranging the starting positions of row data in point cloud data to maintain horizontal and vertical alignment, the method addresses the issue of missed detections of elongated objects, improving the accuracy of object detection in thinned point cloud data.
Patent Information
- Application Number
- JP2023080618
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-16
- Publication Date
- 2024-11-28
AI Technical Summary
Existing object detection methods fail to accurately detect elongated objects, such as lane separation markers, due to thinning processes that align points vertically, leading to missed detections.
The method rearranges the starting positions of row data in point cloud data to ensure horizontal and vertical alignment of distance measurement points, allowing for accurate detection of objects by adjusting the column and row positions of data acquisition, thereby maintaining the integrity of elongated object detection.
This approach effectively suppresses the failure to detect elongated objects by ensuring that position information is acquired from strategically positioned points, enhancing the accuracy of object detection in thinned point cloud data.
Smart Images

Figure 2024164902000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an object detection method and an object detection device. [Background technology]
[0002] An obstacle detection device is known that obtains coordinates of a point that represents a part of an object in a coordinate system consisting of an X-axis extending in the vehicle width direction, a Y-axis that is perpendicular to the X-axis and runs along the road surface on which the vehicle is located, and a Z-axis that is perpendicular to the X-axis and Y-axis, thins out some of the obtained points, derives a surface normal vector for each point remaining after the thinning, and extracts, as an obstacle, a point cloud that is a collection of points whose angle of the surface normal vector with respect to the XY plane is less than a first threshold value, and whose number of points is equal to or greater than a second threshold value (Patent Document 1). [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent Publication No. 2021-85828 Summary of the Invention [Problem to be solved by the invention]
[0004] In the above-mentioned conventional technology, when some points are thinned out from the acquired points, the points remaining after thinning are thinned out so that they are aligned in the Z-axis direction (up-down direction), for example, by setting the distance between adjacent points in the X-axis direction (left-right direction) to a specified distance or more, resulting in a problem that objects that are long and thin in the vertical direction, such as lane separators (guide posts) installed on the road surface, cannot be properly detected.
[0005] An object detection method and an object detection device that can reduce the risk of missing an object when detecting an object from thinned point cloud data. [Means for solving the problem]
[0006] The present invention solves the above problem by setting a start position for starting acquisition of position information in first row data in which position information is arranged in the left-right and up-down directions of a moving body, second row data directly above and below the first row data, and third row data directly above and below the first row data, in point cloud data in which position information of a target object's ranging points is arranged two-dimensionally in the left-right and up-down directions of a moving body, such that the left-right column position of the first start position of the first row data is different from the left-right column position of the second start position of the second row data and the left-right column position of the third start position of the third row data. Effect of the Invention
[0007] According to the present invention, it is possible to suppress overlooking of detection of an object when detecting the object from thinned point cloud data. [Brief description of the drawings]
[0008] [Figure 1] 1 is a block diagram showing an example of an embodiment of an object detection device according to the present invention. [Diagram 2] FIG. 2 is a plan view illustrating an example of the data generating device of FIG. [Diagram 3] 3 is a front view showing an example of point cloud data generated by the data generating device of FIG. 2. [Figure 4] FIG. 4 is a front view showing an example of setting a start position in the point cloud data of FIG. [Diagram 5] 4 is a front view showing another example of setting the start position in the point cloud data of FIG. 3. FIG. [Figure 6] 7 is a front view showing yet another example of setting a start position in the point cloud data of FIG. 3. FIG. [Figure 7] 7 is a front view showing yet another example of setting a start position in the point cloud data of FIG. 3. FIG. [Figure 8] 2 is a flowchart showing an example of a processing procedure in the object detection device of FIG. 1. [Figure 9] 10 is a flowchart showing another example of the processing procedure in the object detection device of FIG. [Figure 10]4 is a front view showing an example of a start position set by an object detection device according to a comparative example of the present invention in the point cloud data of FIG. 3. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0010] [Configuration of object detection device] FIG. 1 is a block diagram showing an object detection device 1 according to an embodiment of the present invention. The object detection device 1 is a device that detects objects present around a moving object such as a vehicle. A moving object is an object that moves according to a user's operation or instruction, and includes a vehicle, a mobile robot, and the like. An object is a detection target of the object detection device, and is not particularly limited, but is, for example, an object used to recognize the current position of the moving object. Specific examples of the object include boundary lines, center lines, road markings, guard rails, curbs, road signs, traffic lights, and crosswalks. In addition, the object also includes other vehicles, motorcycles, bicycles, buildings around the moving object, and roadside installations (mailboxes, vending machines, and the like).
[0011] 1, the object detection device 1 includes a data generating device 10, map information 20, and a control device 30. These devices are connected by a CAN (Controller Area Network) or other in-vehicle LAN, and can exchange information with each other.
[0012] The data generating device 10 is a device provided on a moving body, and generates point cloud data in which position information of distance measuring points of an object is two-dimensionally arranged in the left-right and up-down directions of the moving body, and specific examples thereof include a laser radar, a millimeter wave radar, and a LiDAR (light detection and ranging) unit. Providing the data generating device 10 on a moving body includes mounting the data generating device 10 on the moving body, fixing the data generating device 10 to the moving body, loading the data generating device 10 on the moving body, and the like. Providing the data generating device 10 on a moving body also includes attaching the data generating device 10 to a human being.
[0013] The distance measurement point of the object is a point on the object whose distance to the data generating device 10 is measured by the data generating device 10, and the position information of the distance measurement point is the coordinates of the distance measurement point, the distance from the data generating device 10 to the distance measurement point, etc. The data generating device 10 generates point cloud data by scanning electromagnetic waves in the left-right direction of the moving object. For example, if the moving object is a vehicle, the data generating device 10 detects the distance to the distance measurement point and the direction (angle) of the distance measurement point by irradiating electromagnetic waves along the vehicle width direction and detecting the reflected electromagnetic waves (reflected waves). The irradiation range of the electromagnetic waves is not particularly limited. Examples of the electromagnetic waves include millimeter waves (wavelength: 1 to 10 mm), infrared rays (wavelength: 700 to 1000 nm), and lasers (wavelength: 500 to 800 nm).
[0014] An example of the data generating device 10 is shown using Fig. 2. Fig. 2 is a plan view showing a driving scene in which a vehicle 40, which is a moving body, is driving toward the right side of the drawing (the front side of the vehicle 40). In the driving scene shown in Fig. 2, a lane dividing marker Y and a wall Z, which are objects, are present ahead of the vehicle 40.
[0015] A vehicle 40 shown in FIG. 2 is provided with LiDAR units 10a, 10b, and 10c as data generating devices. The LiDAR unit 10a is fixed to the roof of the vehicle 40, and an irradiation device rotates by a motor to irradiate the entire periphery of the vehicle 40 with a laser. The LiDAR unit 10b is integrated with the windshield of the vehicle 40, and irradiates the laser in the range of 100 to 160° (indicated by angle α in FIG. 2) in the left-right direction of the vehicle 40 and 10 to 30° in the up-down direction. The LiDAR unit 10c is mounted inside the front grille of the vehicle 40, and irradiates the laser in the same range as the LiDAR unit 10b.
[0016] The LiDAR units 10a, 10b, and 10c generate point cloud data in which position information of the distance measurement points where the electromagnetic waves are reflected is arranged in the left-right direction and the up-down direction (two-dimensionally arranged). Specifically, vertical laser pulses of several nanoseconds (for example, 2 to 6 nanoseconds) are scanned in the left-right direction, and the reflected waves of the electromagnetic waves reflected by the objects around the vehicle 40 are detected. Then, the distance to the reflection position is calculated from the time required from the irradiation of the electromagnetic waves to the detection of the reflected waves (the flight time of the laser), and the reflection position (i.e., the position of the distance measurement point) is obtained from the calculated distance and the direction of the irradiation of the electromagnetic waves. The reflection position is obtained, for example, as the coordinates of a three-dimensional orthogonal coordinate system with the position of the LiDAR unit 10b as the origin. By obtaining the reflection position for each pulse, point cloud data in which points indicating the reflection positions of the electromagnetic waves are arranged in the left-right direction and the up-down direction is generated.
[0017] As an example, when the LiDAR unit 10b irradiates a laser pulse 10 times, the laser is reflected at the distance measurement points X1 to X10 shown in FIG. 2, and 10 reflected waves are detected. In this case, the LiDAR unit 10b calculates the flight time for each laser pulse and obtains the distance between the LiDAR unit 10b and the distance measurement points X1 to X10. Then, the coordinates of the distance measurement points X1 to X10 are obtained by triangulation based on the irradiation angle of the laser pulse with respect to the front-rear direction of the vehicle 40. Based on the obtained coordinates, the LiDAR unit 10b generates row data in which the coordinate information of the distance measurement points X1 to X10 is arranged in the left-right direction. In this row data, only the position information of the point corresponding to the distance measurement point X5 among the points constituting the data has position information related to the lane marking Y.
[0018] Returning to Fig. 1, the map information 20 includes road information, facility information, and attribute information thereof, and is high-definition map information that allows, for example, grasping the movement trajectory for each lane. The map information 20 includes, for example, information on road width, road curvature radius, road shoulder structures, road traffic regulations, road junctions and branching points, etc. The map information 20 is stored in a readable state in any recording medium, and is acquired by the control device 30 as necessary.
[0019] The control device 30 is a device for detecting objects around a moving body by controlling and cooperating with the devices constituting the object detection device 1. The control device 30 is, for example, a computer, and includes a CPU (Central Processing Unit) which is a processor, a ROM (Read Only Memory) in which a program is stored, and a RAM (Random Access Memory) which functions as an accessible storage device. The CPU is an operating circuit for executing the programs stored in the ROM and realizing the functions of the control device 30.
[0020] The control device 30 has a detection function for detecting an object. A program for implementing the detection function is stored in the ROM, and the detection function is implemented by the CPU executing the program stored in the ROM. In Fig. 1, a setting unit 31, an acquisition unit 32, a detection unit 33, and an estimation unit 34 are illustrated as functional blocks for implementing the detection function for the sake of convenience. The functions of each functional block will be described below with reference to Fig. 3.
[0021] [Functions of each function block] Fig. 3 is a front view showing an example of point cloud data 50 generated by the LiDAR unit 10b shown in Fig. 2 in the driving scene shown in Fig. 2. The LiDAR unit 10b generates point cloud data 50 in which position information of the reflection positions of the laser (i.e., the positions of the ranging points) is two-dimensionally arranged in the left-right and up-down directions of the moving body. For example, the point cloud data 50 shown in Fig. 3 has 15 columns in the up-down direction, in which 21 points corresponding to the position information of the ranging points are arranged (arranged) in the left-right direction. For the sake of explanation, the wall Z is omitted from Fig. 3.
[0022] Since the data generating device 10 scans electromagnetic waves in the left-right direction of the moving object, the point cloud data 50 includes at least one row data, but the points constituting the row data do not necessarily have to correspond in the up-down direction as shown in FIG. 3, and the left-right positions (hereinafter also referred to as column positions) of the points constituting the point cloud data 50 do not necessarily have to be aligned. In other words, the left-right intervals of the ranging points do not have to be equal, and there does not have to be another ranging point above and / or below any ranging point. In addition, the vertical intervals between the row data do not have to be equal, and the vertical intervals between the row data (i.e., the vertical intervals between the ranging points) can be set to an appropriate value within a range in which the target object can be appropriately detected.
[0023] In the point cloud data 50 shown in Fig. 3, the data rows are numbered A to O in order from the top, and are referred to as row data A to row data O, respectively. Also, the points indicating the position information constituting each row data are numbered 1 to 21 in order from the left, and each point constituting the point cloud data 50 is represented by a combination of a code and a number. For example, the fifth point from the top and third point from the left is the third point of row data E, and is therefore represented as "point (E, 3)".
[0024] For example, when detecting an object from the point cloud data 50 shown in Fig. 3, if it takes a long time to process the information, the points in the point cloud data 50 are thinned out to reduce the number of points (i.e., the amount of position information of the ranging points) before detecting the object. As an example, when thinning out the points in the point cloud data 50 by acquiring position information of ranging points every fourth point (every fifth point) from the row data, position information is acquired from the 1st, 6th, 11th, 16th, and 21st points of each row data, as shown in Fig. 10 according to a comparative example of the present invention.
[0025] In this case, the hatched points for which position information has been acquired are arranged vertically, so that, for example, the position information of the 7th to 10th points in each row of data is not used to detect the target object. In the point cloud data 50 shown in Fig. 3, if it is assumed that points (B, 7) to (O, 7) and points (B, 8) to (O, 8) among the points constituting the point cloud data 50 indicate the positions of the distance measurement points on the lane dividing mark Y and the other points indicate the positions of the distance measurement points on the wall Z, the position information of the lane dividing mark Y cannot be acquired and the lane dividing mark Y cannot be detected by the thinning process shown in Fig. 10.
[0026] Therefore, the control device 30 of this embodiment uses the functions of the setting unit 31, the acquiring unit 32, and the detecting unit 33 to thin out the points of the point cloud data 50 shown in FIG. 3 through a thinning process shown in FIG.
[0027] The setting unit 31 has a function of setting a start position for starting acquisition of position information of distance measurement points in a first row of data in which points corresponding to position information of distance measurement points are aligned (arranged) in the left-right direction, a second row of data immediately above the first row of data, and a third row of data immediately below the first row of data, in the point cloud data 50 acquired from the data generating device 10. The control device 30 sets a column position in the left-right direction and a row position in the up-down direction as the start position. In the example shown in Fig. 4, the column positions are indicated by numbers 1 to 21, and the row positions are indicated by symbols A to O.
[0028] Furthermore, the control device 30 sets a position in the vertical direction that does not correspond to the second start position of the second row data and the third start position of the third row data as the first start position of the first row data. In other words, the column position of the first start position of the first row data is different from the column position of the second start position of the second row data and the column position of the third start position of the third row data. As long as this condition is satisfied, the control device 30 can set any position (point) of the first row data, the second row data, and the third row data as the first start position, the second start position, and the third start position, respectively.
[0029] Fig. 4 is a front view showing an example of the start position set by the control device 30 for the point cloud data 50 shown in Fig. 3. In the example shown in Fig. 4, the control device 30 repeatedly executes a process of setting the second start position to a position (point) one position (point) to the left of the position (point) corresponding to the first start position in the vertical direction, and setting the third start position to a position (point) one position (point) to the right of the position (point) corresponding to the first start position in the vertical direction, thereby setting the start position of each line data. Specifically, point (A, 1), point (B, 2), point (C, 3), point (D, 4), point (E, 5), point (F, 6), point (G, 7), point (H, 8), point (I, 9), point (J, 10), point (K, 11), point (L, 12), point (M, 13), point (N, 14) and point (O, 15) are set as the start position of each line data.
[0030] In the example shown in FIG. 4, for example, if row data B is the first row data, row data A immediately above row data B is the second row data, and row data C immediately below row data B is the third row data. Point (B, 2) is the first start position, point (A, 1) is the second start position, and point (C, 3) is the third start position. In this case, point (B, 2) is not point (B, 1) corresponding vertically to point (A, 1), nor point (B, 3) corresponding vertically to point (C, 3). In addition, the column position of point (A, 1) is one position to the left of the column position of point (A, 2) corresponding vertically to point (B, 2), and the column position of point (C, 3) is one position to the right of the column position of point (C, 2) corresponding vertically to point (B, 2).
[0031] The acquisition unit 32 has a function of acquiring position information of ranging points at predetermined intervals from the start position set by the function of the setting unit 31. The control device 30 acquires position information of ranging points at predetermined intervals from the first start position, the second start position, and the third start position in each of the first line data, the second line data, and the third line data. The predetermined number can be set to an appropriate value depending on the number of line data and the number of points constituting the line data (i.e., the number of pieces of position information).
[0032] When acquiring position information every predetermined number of points, position information is acquired every predetermined number of points in at least one of the right and left directions from the start position. For example, when point (E, 5) shown in FIG. 4 is set as the start position and position information is acquired every third point (every fourth point) in row data E, position information of points (E, 5) and (E, 1) is acquired (left direction), position information of points (E, 5), (E, 9), (E, 13), (E, 17), and (E, 21) is acquired (right direction), or position information of points (E, 1), (E, 5), (E, 9), (E, 13), (E, 17), and (E, 21) is acquired (left and right directions). Note that it is not necessary to acquire position information of the start position.
[0033] In the example shown in Fig. 4, the control device 30 acquires position information every four points (every five points) to the right from the above-mentioned 15 starting positions. In this case, the control device 30 acquires position information from the hatched points. The hatched points from which position information is acquired include point (B, 7), point (C, 8), point (G, 7), and point (H, 8), so that points indicating the positions of the distance measurement points on the lane marking Y remain in the thinned point cloud data. Hereinafter, points from which position information is acquired in the point cloud data 50 will be shown hatched.
[0034] The detection unit 33 has a function of detecting an object existing around the moving body from the position information acquired by the function of the acquisition unit 32. The control device 30 detects an object existing around the moving body from the position information of the ranging points (coordinates of the ranging points, distance to the object, etc.). For example, the control device 30 creates an image in which the ranging points are plotted on a three-dimensional map based on the position information of the ranging points, and detects the object from the image.
[0035] 4, the control device 30 generates an image on which the distance measurement points are plotted from the position information of the distance measurement points acquired by the function of the acquisition unit 32, and performs processing such as pattern matching on the image to detect the target object. Since the control device 30 uses the position information of points (B, 7), (C, 8), (G, 7), and (H, 8) when generating the image, it is possible to detect the lane dividing marker Y present in front of the vehicle 40.
[0036] Next, Fig. 5 is a front view showing another example of the start position set by the control device 30 for the point cloud data 50 shown in Fig. 3. Fig. 5 shows an example of setting the start position when acquiring position information every 9 points (every 10 points) to the right, and points (A, 1), (B, 3), (C, 5), (D, 7), (E, 9), (F, 11), (G, 13), (H, 15), (I, 17), (J, 19) and (K, 21) are set as the start positions.
[0037] In the example shown in FIG. 5, for example, if the row data B is the first row data, the row data A immediately above the row data B is the second row data, and the row data C immediately below the row data B is the third row data. Then, point (B, 3) is the first start position, point (A, 1) is the second start position, and point (C, 5) is the third start position. In this case, the difference between the start positions in the left-right direction is set larger than in the example shown in FIG. 4, and thus the points from which position information is obtained can be set more sparsely compared to the example shown in FIG. 4. In this way, the control device 30 may set at least one of the difference between the left-right direction between the first start position and the second start position and the difference between the left-right direction between the first start position and the third start position to be larger as the predetermined number increases. The difference between the start positions may be set by the distance between the start positions, or may be set by the interval between the points of the row data, the interval of the irradiation angle of the electromagnetic wave of the data generating device, or the like.
[0038] 5, the points for which position information is acquired include point (D,7), so that points indicating the positions of the distance measurement points on the lane dividing marker Y remain in the thinned point cloud data. As a result, in the example shown in FIG. 5, the control device 30 can detect the lane dividing marker Y that exists ahead of the vehicle 40.
[0039] In the example shown in Fig. 4, a start position is set for each line data, and position information is acquired for each line data at a predetermined number of positions from the start position, but it is not necessary to acquire position information from all line data as in the example shown in Fig. 5. For example, it is not necessary to acquire position information from line data in a portion where the probability of an object being present is low, such as the upper side of the point cloud data 50.
[0040] Next, Fig. 6 is a front view showing yet another example of the start position set by the control device 30 for the point cloud data 50 shown in Fig. 3. In the example shown in Fig. 6, the control device 30 repeatedly executes a process of setting the column positions of the second start position and the third start position to the position one position to the left of the column position of the first start position, and sets the start position of each row data. Specifically, point (A,1), point (B,2), point (C,1), point (D,2), point (E,1), point (F,2), point (G,1), point (H,2), point (I,1), point (J,2), point (K,1), point (L,2), point (M,1), point (N,2) and point (O,1) are set as the start position of each row data.
[0041] 6, for example, if row data B is the first row data, row data A immediately above row data B is the second row data, and row data C immediately below row data B is the third row data. Point (B,2) is the first start position, point (A,1) is the second start position, and point (C,1) is the third start position. In this case, the column positions of points (A,1) and (C,1) are one position to the left of the column positions of points (A,2) and (C,2) that correspond vertically to point (B,2).
[0042] Moreover, the control device 30 can set the same start position as the start position shown in FIG. 6 by repeatedly executing a process of setting the column positions of the second start position and the third start position to positions one position to the right of the column position of the first start position. For example, if row data C is the first row data, row data B immediately above row data C is the second row data, and row data D immediately below row data C is the third row data. Then, point (C, 1) is the first start position, point (B, 2) is the second start position, and point (D, 2) is the third start position. In this case, the column positions of points (B, 2) and (D, 2) are positions one position to the right of the column positions of points (B, 1) and (D, 1) that correspond to point (C, 1) in the vertical direction.
[0043] 6, the control device 30 acquires position information for every 4th point (every 5th point) from the start position 15 described above in a rightward direction. In this case, the control device 30 acquires position information from the hatched points. The points from which position information is acquired include point (B, 7), point (D, 7), point (F, 7), point (H, 7), point (J, 7), point (L, 7), and point (N, 7), so that the points indicating the positions of the distance measurement points on the lane dividing marker Y remain in the thinned point cloud data.
[0044] 4, the control device 30 generates an image in which the reflection positions are plotted from the acquired position information, and performs processing such as pattern matching on the image to detect the target object. When generating the image, the control device 30 uses the position information of points (B, 7), (D, 7), (F, 7), (H, 7), (J, 7), (L, 7), and (N, 7), and therefore can detect the lane dividing marker Y present in front of the vehicle 40.
[0045] For positions where the irradiated electromagnetic waves are not reflected, such as the sky, or positions beyond the detection limit distance of the data generating device 10, position information of the ranging points cannot be acquired. In the example shown in Fig. 6, it is assumed that the reflected waves of the laser could not be detected at the ranging points corresponding to points (B, 9), (B, 10), (C, 10), and (C, 11) shown by the dashed lines, and therefore position information of the ranging points could not be acquired. When such points for which position information cannot be acquired exist in the row data, the control device 30 may acquire position information every predetermined number of points, assuming that there are no points for which position information cannot be acquired.
[0046] In the example shown in Fig. 6, position information would normally be obtained from points (B,2), (B,7), (B,12), and (B,17) of row data B, but since position information cannot be obtained for points (B,9) and (B,10), position information is obtained from points (B,2), (B,7), (B,14), and (B,19) of row data B, excluding (skipping) these two points. Similarly, position information would normally be obtained from points (C,1), (C,6), (C,11), (C,16), and (C,21) of row data C, but since position information cannot be obtained for points (C,10) and (C,11), position information is obtained from points (C,1), (C,6), (C,13), and (C,18) of row data C, excluding (skipping) these two points.
[0047] Next, Fig. 7 is a front view showing yet another example of the start positions set by the control device 30 for the point cloud data 50 shown in Fig. 3. Fig. 7 shows an example of setting the start positions when acquiring position information every four points (every five points) in the right direction, and points (A, 8), (B, 1), (C, 9), (D, 2), (E, 10), (F, 3), (G, 11), (H, 4), (I, 12), (J, 5), (K, 13), (L, 6), (M, 14), (N, 7), and (O, 15) are set as the start positions.
[0048] In the example shown in FIG. 7, for example, if row data B is the first row data, row data A immediately above row data B is the second row data, and row data C immediately below row data B is the third row data. Point (B, 1) is the first start position, point (A, 8) is the second start position, and point (C, 9) is the third start position. In this case, if row data D different from row data A to C is the fourth row data, the fourth start position from which acquisition of position information starts in the fourth row data is point (D, 2), and the difference in the left-right direction between point (B, 1) and point (D, 2) is smaller than the difference in the left-right direction between point (B, 1) and point (A, 8) and the difference in the left-right direction between point (B, 1) and point (C, 9). This allows the points from which position information is acquired to be set more sparsely compared to the example shown in FIG. 4. In this way, the control device 30 may set a fourth start position from which to start acquiring position information in fourth line data different from the first line data, the second line data, and the third line data, and set the left-right difference between the first start position and the fourth start position to be smaller than either one of the left-right difference between the first start position and the second start position and the left-right difference between the first start position and the third start position.
[0049] 7, the points for which position information is acquired include point (D, 7), point (F, 8), and point (N, 7), so points indicating the positions of the distance measurement points on the lane dividing marker Y remain in the thinned point cloud data. As a result, in the example shown in FIG. 7, the control device 30 can detect the lane dividing marker Y that is present ahead of the vehicle 40.
[0050] The control device 30 may randomly set a start position for starting acquisition of position information for each row data of the point cloud data 50. In this case, it is determined whether or not the column position of the randomly set first start position is different from the column position of the second start position set in the row data immediately above and the column position of the third start position set in the row data immediately below, and if the column position of the first start position is the same as at least one of the column positions of the second start position and the third start position, the first start position is set again.
[0051] The control device 30 may set row data numbers for multiple row data in order from top to bottom, calculate the remainder when the sum of the total number of row data and the row data number is divided by a first integer that is smaller than the total number of row data and greater than 1 for each row data, perform a process to convert the calculated remainder into an integer to obtain a second integer, and set the position that is the second integer number from the position of the leftmost or rightmost point for each row data as the start position. The first integer may be set to an appropriate integer. The process to convert the calculated remainder into an integer is a process to round off, round up, or round down to one decimal place.
[0052] Returning to Fig. 1, the estimation unit 34 has a function of estimating the current position of the moving body using information on the position of the object detected by the function of the detection unit 33. The control device 30 judges whether or not a characteristic part of the object detected from the point cloud data 50 matches a characteristic part of the object registered in advance in the map information 20, and if it is judged that the two characteristic parts match, it estimates the current position of the moving body from the position of the object registered in the map information 20 and the relative position of the object with respect to the moving body detected by the function of the detection unit 33. On the other hand, if it is judged that the two characteristic parts do not match, the current position of the moving body is not estimated.
[0053] For example, when the lane separation mark Y and wall Z shown in FIG. 2 are detected from the point cloud data 50 shown in FIG. 3, the control device 30 determines the characteristic parts of the lane separation mark Y and wall Z and acquires the characteristic parts of the object near the acquisition position on the map information 20. The acquisition position is, for example, the position last detected as the traveling position of the vehicle 40, and the characteristic parts are, for example, the characteristic lines of the shapes of the lane separation mark Y and wall Z. When it is determined that the shapes of the detected lane separation mark Y and wall Z match the shapes of the lane separation mark Y and wall Z acquired from the map information 20, the current position of the vehicle 40 is estimated from the positions of the lane separation mark Y and wall Z registered in the map information 20 and the relative positions of the lane separation mark Y and wall Z with respect to the vehicle 40. On the other hand, when it is determined that the shapes of the detected lane separation mark Y and wall Z do not match the shapes of the lane separation mark Y and wall Z acquired from the map information 20, the estimation of the current position of the vehicle 40 is repeated or the estimation of the current position of the vehicle 40 is stopped.
[0054] [Processing in object detection device] The procedure for information processing by the control device 30 will be described with reference to Figs. 8 and 9. Fig. 8 is an example of a flowchart showing information processing executed in the object detection device 1 of the present embodiment. The processing described below is executed by a processor (CPU) included in the control device 30.
[0055] First, in step S1, the setting unit 31 functions to acquire point cloud data 50 from the data generating device 10, and in the subsequent step S2, a start position is set in the first to third lines of data of the point cloud data 50. In step S3, the acquisition unit 32 functions to acquire position information at the set start position, and in the subsequent step S4, position information is acquired at predetermined intervals from the start position. In step S5, the detection unit 33 functions to detect surrounding objects from the acquired position information. Thereafter, execution of the routine ends.
[0056] 9 is a flowchart showing another example of information processing executed in the object detection device 1 of the present embodiment. The processing described below is executed by a processor (CPU) included in the control device 30.
[0057] First, in step S11, surrounding environment information is acquired by the function of the estimation unit 34, and it is determined whether or not the scene is one in which the current position of a moving object should be estimated using the detection result of the data generating device 10. If it is determined that the scene is not one in which the current position of a moving object should be estimated using the detection result of the data generating device 10, the determination of step S11 is repeated. On the other hand, if it is determined that the scene is one in which the current position of a moving object should be estimated using the detection result of the data generating device 10, the process proceeds to step S12.
[0058] In step S12, the setting unit 31 functions to acquire point cloud data 50 from the data generating device 10, and in the following step S13, a start position is set for each line data of the point cloud data 50. In step S14, the acquisition unit 32 functions to acquire position information at the set start position, and in the following step S15, it is determined whether or not there is a point in the line data for which position information cannot be acquired. If it is determined that there is no point for which position information cannot be acquired, the process proceeds to step S16, where position information is acquired for each line data at a predetermined number of points from the start position. In contrast, if it is determined that there is a point for which position information cannot be acquired, the process proceeds to step S17, where position information is acquired for each line data at a predetermined number of points from the start position, excluding points for which position information cannot be acquired (skipping or assuming that they do not exist).
[0059] In step S18, the detection unit 33 detects surrounding objects from the acquired position information. In step S19, the estimation unit 34 determines the characteristic parts of the detected objects, and in the following step S20, the characteristic parts of the objects are acquired from the point cloud data 50 of the map information 20, and in the following step S21, the characteristic parts are compared with each other. In step S22, it is determined whether the characteristic parts match with each other. If it is determined that the characteristic parts match with each other, the process proceeds to step S23, where the current position of the moving object is estimated from the position of the object on the map information 20 and the relative position of the detected object with respect to the moving object. On the other hand, if it is determined that the characteristic parts do not match with each other, the process proceeds to step S19.
[0060] [Embodiments of the present invention] As described above, according to the present embodiment, in an object detection method that is executed by an object detection device 1 that includes a data generating device 10 that is provided on a moving body and generates point cloud data 50 in which position information of distance measuring points of an object is two-dimensionally arranged in the left-right direction and the up-down direction of the moving body, the object detection device 1 generates the position information in a first row of data in which the position information is arranged in the left-right direction, a second row of data immediately above the first row of data in the up-down direction, and a third row of data immediately below the first row of data in the up-down direction in the point cloud data 50 acquired from the data generating device 10. The present invention provides an object detection method that sets a start position for starting acquisition of information, a horizontal column position of a first start position of the first row data being different from a horizontal column position of a second start position of the second row data and a horizontal column position of a third start position of the third row data, and acquires position information from the first start position, the second start position and the third start position at predetermined intervals in the first row data, the second row data and the third row data, and detects objects existing around the moving body from the position information acquired at predetermined intervals. This makes it possible to suppress oversight of detection of objects when detecting objects from thinned point cloud data 50.
[0061] According to the object detection method of the present embodiment, the object detection device 1 sets at least one of the difference in the left-right direction between the first start position and the second start position and the difference in the left-right direction between the first start position and the third start position to be larger as the predetermined number increases, thereby making it possible to further reduce missed detection of the object.
[0062] According to the object detection method of the present embodiment, the object detection device 1 sets a fourth start position for starting acquisition of the position information in fourth line data different from the first line data, the second line data, and the third line data, and sets the difference in the left-right direction between the first start position and the fourth start position to be smaller than either one of the difference in the left-right direction between the first start position and the second start position and the difference in the left-right direction between the first start position and the third start position. This makes it possible to further reduce missed detection of the object.
[0063] According to the object detection method of this embodiment, the object detection device 1 sets the column position in the left-right direction of the second start position to a position one position to the left of the column position in the left-right direction of the first start position, and sets the column position in the left-right direction of the third start position to a position one position to the right of the column position in the left-right direction of the first start position. This makes it possible to set points for obtaining position information more sparsely.
[0064] According to the object detection method of the present embodiment, the object detection device 1 sets the column positions in the left-right direction of the second start position and the third start position to a position one position to the left of the column position in the left-right direction of the first start position, or sets the column positions in the left-right direction of the second start position and the third start position to a position one position to the right of the column position in the left-right direction of the first start position. This makes it possible to set points from which position information is obtained more sparsely.
[0065] According to the object detection method of the present embodiment, when there are ranging points for which the position information cannot be acquired, the object detection device 1 acquires the position information for every predetermined number of ranging points excluding the ranging points for which the position information cannot be acquired. This makes it possible to further reduce missed detection of the object.
[0066] Further, according to this embodiment, there is provided a data generating device 10 that is provided on a moving body and generates point cloud data 50 in which position information of distance measuring points of an object is two-dimensionally arranged in the left-right direction and the up-down direction of the moving body, and a setting unit 31 that sets a start position for starting acquisition of the position information in a first row of data in which the position information is arranged in the left-right direction, a second row of data immediately above the first row of data in the up-down direction, and a third row of data immediately below the first row of data in the up-down direction in the point cloud data 50 acquired from the data generating device 10, and an object detection device 1 including: a setting unit 31 in which the column position in the left-right direction of the second start position of the second row data is different from the column position in the left-right direction of the second start position of the second row data and the column position in the left-right direction of the third start position of the third row data, an acquisition unit 32 that acquires the position information for every predetermined number of positions from the first start position, the second start position, and the third start position set by the setting unit 31 in the first row data, the second row data, and the third row data, and a detection unit 33 that detects objects existing around the moving body from the position information acquired by the acquisition unit 32. This makes it possible to suppress oversight of detection of objects when detecting objects from thinned point cloud data 50. [Explanation of symbols]
[0067] 1...Object detection device 10...Data generation device 10a, 10b, 10c…LiDAR unit 20…Map information 30...Control device 31…Settings section 32…Acquisition part 33…Detection section 34…Estimation part 40…Vehicle 50...Point cloud data X1,X2,X3,X4,X5,X6,X7,X8,X9,X10... Focusing points Y... Lane separation sign (guide post) Z…Wall
Claims
1. An object detection method executed by an object detection device including a data generating device provided on a moving body and generating point cloud data in which position information of distance measuring points of an object is two-dimensionally arranged in a left-right direction and a top-bottom direction of the moving body, The object detection device includes: in the point cloud data acquired from the data generating device, a start position for starting acquisition of the position information is set for first row data in which the position information is arranged in the left-right direction, second row data immediately above the first row data in the up-down direction, and third row data immediately below the first row data in the up-down direction, and a column position in the left-right direction of a first start position of the first row data is different from a column position in the left-right direction of a second start position of the second row data and a column position in the left-right direction of a third start position of the third row data, acquiring the position information for each predetermined number of lines from the first start position, the second start position, and the third start position in the first line data, the second line data, and the third line data; An object detection method for detecting the target object existing around the moving body from the position information acquired at every predetermined number of pieces of position information.
2. The object detection method according to claim 1 , wherein the object detection device sets at least one of the left-right difference between the first start position and the second start position and the left-right difference between the first start position and the third start position to be larger as the specified number increases.
3. The object detection device includes: a fourth start position for starting acquisition of the position information is set in fourth line data different from the first line data, the second line data, and the third line data; 2. The object detection method according to claim 1, wherein the left-right difference between the first start position and the fourth start position is set to be smaller than either the left-right difference between the first start position and the second start position or the left-right difference between the first start position and the third start position.
4. The object detection device includes: A column position of the second start position in the left-right direction is set to a position one position to the left of a column position of the first start position in the left-right direction; The object detection method according to claim 1 , wherein the third start position is set to a position immediately to the right of the first start position in the left-right direction.
5. The object detection device includes: Set the column positions of the second start position and the third start position in the left-right direction to a position immediately to the left of the column position of the first start position in the left-right direction, or The object detection method according to claim 1 , wherein the second start position and the third start position are set to positions immediately to the right of the first start position in the left-right direction.
6. The object detection method according to any one of claims 1 to 5, wherein, when there are ranging points for which the position information cannot be obtained, the object detection device obtains the position information every predetermined number of ranging points excluding the ranging points for which the position information cannot be obtained.
7. a data generating device provided on the moving body for generating point cloud data in which position information of distance measuring points of an object is two-dimensionally arranged in the left-right direction and the up-down direction of the moving body; a setting unit that sets a start position for starting acquisition of position information on first row data, in which the position information is arranged in the left-right direction, second row data immediately above the first row data in the up-down direction, and third row data immediately below the first row data in the up-down direction, in the point cloud data acquired from the data generating device, wherein a column position in the left-right direction of a first start position of the first row data is different from a column position in the left-right direction of a second start position of the second row data and a column position in the left-right direction of a third start position of the third row data; an acquisition unit that acquires the position information at predetermined intervals from the first start position, the second start position, and the third start position set by the setting unit in the first line data, the second line data, and the third line data; a detection unit that detects an object existing around the moving body from the position information acquired by the acquisition unit.
Citation Information
Patent Citations
Image pickup device
JP2004104516A
Image processor, image display apparatus, image processing method, program for allowing computer to perform the method, and recording medium
JP2006259403A
Method for accelerating three-dimensional object segmentation with point cloud simplifications
US20210295522A1
Processing device and point group curtailment method
WO2021065371A1
Obstacle detector
JP2021085828A
