JP7737776B2 - Vehicle position detection method and position detection device - Google Patents

Description

The present invention relates to a method and device for detecting the position of a detected object, and in particular to a position detection technology that is suitable for use when the detected object is a vehicle.

For example, one common method for efficiently transporting finished vehicles such as automobiles mass-produced in a factory to a specified location such as a shipping point is to load multiple finished vehicles into a loading vessel such as a container, transport the loading vessel by land to a port near the destination in a transport vehicle such as a large truck, and then transfer the loading vessel to a transport vessel for water transport to the destination (see, for example, Patent Document 1).

Patent Document 2 also proposes a system in which a transport vehicle, which uses an unmanned, automatic towing vehicle to tow multiple carts along a predetermined route within a limited area, transports the completed vehicle to be transported, loaded onto the cart, to its destination.

Japanese Patent Application Laid-Open No. 2004-123258 Japanese Patent Application Laid-Open No. 2019-36036 International Publication No. 2013/31418

In recent years, the manufacturing industry has also seen an increasing need to take measures to compensate for labor shortages in preparation for the coming aging society. From this perspective, when considering how to transport completed vehicles efficiently with a small number of people, one possible method would be to make the transport vehicles self-propelled, as described in Patent Document 2, and then manufacture them at low cost and prepare a large number of them, so that each completed vehicle can be loaded onto the transport vehicle and transported automatically.

On the other hand, if we consider loading completed vehicles onto guided vehicles for automatic transport as described above, it is important to know the exact position of the completed vehicle relative to the guided vehicle in order to quickly approach the completed vehicle.

When focusing on points on a completed vehicle's characteristic outer shape, one possible method for detecting the position of a completed vehicle is to use technology such as that described in Patent Document 3, which applies a Hough transform to points in an image containing position coordinate data to detect straight lines contained in the image. In other words, since the body of a completed vehicle (particularly the front part of the body) generally has a shape close to a box, it would appear that by, for example, acquiring position coordinate data (point cloud data) on the body surface using an appropriate sensor and then applying a Hough transform to the acquired position coordinate data, it would be possible to detect straight lines along the sides and front of the body as the body outline from an image containing position coordinate data as points.

However, with this method, a huge amount of position coordinate data must be acquired by the sensor and a Hough transform must be applied to all of the acquired position coordinate data, which means that the calculations involved in the Hough transform require a significant amount of time. As mentioned above, in order to quickly approach the completed vehicle, it is necessary to constantly recognize the relative position of the completed vehicle in relation to the transport vehicle and have the transport vehicle approach the completed vehicle in the appropriate direction and at the appropriate speed. However, if it takes time to detect the position of the completed vehicle, it is difficult to have the transport vehicle approach the completed vehicle smoothly and quickly.

The above-mentioned problem is not limited to when a guided vehicle is approached by a vehicle, but can occur in any situation where the position of a target object needs to be detected in a short period of time.

In light of the above, the technical problem to be solved in this specification is to enable the position of a target object to be detected in a short period of time by using a Hough transform on point cloud data related to position information.

The above-mentioned problem is solved by the position detection method for a detectable object according to the present invention. Specifically, this position detection method is characterized by comprising: a point cloud data acquisition process for acquiring primary point cloud data relating to position information on the surface of the detectable object; a voxel generation process for generating multiple voxels based on the primary point cloud data; a data number reduction process for reducing the number of primary point cloud data located within each voxel; a provisional outline detection process for detecting a provisional outline of the detectable object by applying a Hough transform to the secondary point cloud data obtained in the data number reduction process; a final outline detection process for detecting the final outline of the detectable object by applying a Hough transform to primary point cloud data located around the provisional outline, among the primary point cloud data; and a position detection process for detecting the position of the detectable object based on the final outline.

In this way, the position detection method of the present invention generates multiple voxels based on the acquired primary point cloud data, reduces the number of primary point cloud data located within each voxel, and performs a Hough transform on the secondary point cloud data, which is the reduced primary point cloud data, to detect a tentative outline of the object to be detected. This reduces the amount of point cloud data used in the Hough transform, shortening the time required to detect the tentative outline. Meanwhile, the position of the object to be detected is detected by detecting the actual outline using a Hough transform on the original primary point cloud data, allowing for highly accurate position detection. Furthermore, by limiting the primary point cloud data used during position detection to the area surrounding the tentative outline, only point cloud data that is likely to be relevant to detecting the actual outline can be used. This makes it possible to efficiently detect the position of the object to a short period of time while still achieving high position detection accuracy.

The position detection method according to the present invention may further include a reliability evaluation step of evaluating the reliability of the outline. In this case, the reliability evaluation step may obtain the proportion of voxels containing secondary point cloud data for at least the voxels through which the outline passes, and evaluate the outline to be more reliable the higher the proportion.

In this way, by providing a reliability evaluation process, it is possible to increase the reliability of the main outline detected in the main outline detection process. This also makes it possible to increase the reliability of the position of the object detected based on the main outline. Furthermore, by obtaining the proportion of voxels containing secondary point cloud data for at least the voxels passed by the main outline and evaluating the reliability of the main outline based on this proportion, it is possible to efficiently and quickly evaluate the reliability of the main outline in this process as well.

In addition, in the position detection method according to the present invention, the reliability evaluation step may obtain the proportion of voxels for which secondary point cloud data exists for voxels adjacent to voxels through which the outline passes on the side farther from the detected object, and evaluate the outline to be more reliable the lower the proportion.

In this way, by evaluating the reliability of the outline based on the proportion of voxels containing secondary point cloud data, not only for the voxels through which the outline passes, but also for those voxels adjacent to the voxel on the far side from the detected object, the reliability of the detected outline can be further increased. This makes it possible to further increase the reliability of the position of the detected object detected based on the outline.

Furthermore, in the position detection method according to the present invention, in the reliability evaluation process, the reliability may be evaluated by setting the voxel size to a size larger than the voxels generated in the voxel generation process.

The point cloud data (secondary point cloud data) used in the reliability evaluation process only needs to be the minimum number of data points necessary to evaluate the reliability of the outline, so there is no particular problem in terms of evaluation accuracy even if the voxel size used in the reliability evaluation process is larger than the voxel size used when detecting the provisional outline. Therefore, by increasing the voxel size in this way, it is possible to evaluate the reliability of the outline at the required level, while also being able to evaluate that reliability in a shorter time.

In addition, in the position detection method according to the present invention, the object to be detected may be a vehicle. In this case, the provisional outline detection process may detect, as provisional outlines, provisional side lines that are provisional outlines along the sides of the vehicle and provisional front lines that are provisional outlines that intersect the provisional side lines and are along the front of the vehicle, and the actual outline detection process may detect, as actual outlines, formal side lines that are formal outlines along the sides of the vehicle and formal front lines that are formal outlines that intersect the second side lines and are along the front of the vehicle.

As described above, the present invention enables the position of a detectable object to be detected in a very short time, making it suitable for use when the detection target (detectable object) is a vehicle whose outline is easily detected as a straight line, such as a sideline or frontline. Therefore, by detecting provisional sidelines and provisional frontlines along the sides and front of the vehicle as a provisional outline, and detecting actual sidelines and actual frontlines along the sides and front of the vehicle as an actual outline, it becomes possible to accurately detect the position and posture of the vehicle.

Furthermore, in the position detection method according to the present invention, when a vehicle is used as the detected object, the outline detection process may detect the actual sideline and the actual frontline by excluding the primary point cloud data within the voxel where the intersection of the provisional sideline and the provisional frontline is located from among the multiple voxels.

The sides and front of a vehicle are usually connected via smooth convex curves (corners). Therefore, when point cloud data from the side to the front of a vehicle is acquired using a publicly known point cloud data acquisition sensor such as Lidar, it is conceivable that the point cloud data will be concentrated in the corners between the side and front. Therefore, by excluding primary point cloud data within voxels where the intersections of the provisional sidelines and provisional frontlines are located, it is possible to use only the primary point cloud data that is important for detecting the actual sidelines and the actual frontlines. This improves the accuracy of detecting the actual sidelines and the actual frontlines. Furthermore, excluding some primary point cloud data can shorten the time required for detection.

The above-mentioned problems are also solved by the position detection device for a detectable object according to the present invention. Specifically, this position detection device is characterized by comprising: a point cloud data acquisition unit capable of acquiring primary point cloud data relating to position information on the surface of a detectable object; a voxel generation unit that generates multiple voxels based on the primary point cloud data; a data number reduction unit that reduces the number of primary point cloud data located within each voxel; an outline detection unit that applies a Hough transform to the secondary point cloud data obtained by the data number reduction unit to detect a provisional outline of the detectable object, or that applies a Hough transform to primary point cloud data located around the provisional outline among the primary point cloud data, to detect a real outline of the detectable object; and a position detection unit that detects the position of the detectable object based on this outline.

In this way, the position detection device of the present invention also generates multiple voxels based on the acquired primary point cloud data, reduces the number of primary point cloud data located within each voxel, and performs a Hough transform on the secondary point cloud data, which is the reduced primary point cloud data, to detect a tentative outline of the object to be detected. This reduces the amount of point cloud data used in the Hough transform and shortens the time required to detect the tentative outline. Meanwhile, the position of the object to be detected can be detected by detecting the actual outline using a Hough transform on the original primary point cloud data, allowing for highly accurate position detection. Furthermore, by limiting the primary point cloud data used in this case to the area surrounding the tentative outline, the position of the object to be detected can be detected efficiently and in a short amount of time using only valid point cloud data.

As described above, according to the present invention, it is possible to detect the position of a target object in a short period of time by using a Hough transform on point cloud data related to position information.

1 is a diagram showing the overall configuration of an automatic vehicle transport system according to an embodiment of the present invention; FIG. 2 is a side view of the automatic transport device shown in FIG. 1 . FIG. 2 is a plan view of the automatic transport device shown in FIG. 1 . FIG. 2 is a front view of the automatic transport device shown in FIG. 1 . 1. FIG. 4 is a side view conceptually illustrating an example of acquisition of point cloud data relating to position information of a vehicle surface by a point cloud data acquisition unit (detection unit) provided in the automatic transport device shown in FIG. 2 is a plan view conceptually illustrating an example of acquisition of point cloud data relating to position information of a vehicle surface by a point cloud data acquisition unit provided in the automatic transport device shown in FIG. 1 . FIG. 1 is a diagram illustrating a configuration of a position detection device according to an embodiment of the present invention. 3 is a diagram showing an example of point cloud data relating to position information of a vehicle surface acquired by a point cloud data acquisition unit; FIG. 10 is a diagram conceptually illustrating an example of a distribution pattern of voxels generated by a voxel generating unit. FIG. FIG. 10 is a diagram conceptually illustrating a state after the number of data items in the primary point cloud data has been reduced by the data number reduction unit. 10 is a diagram conceptually showing the positional relationship between a provisional outline detected by an outline detection unit, voxels, and secondary point cloud data. FIG. 10 is a diagram conceptually showing the positional relationship between the outline detected by the outline detection unit, voxels, and primary point cloud data. FIG. FIG. 10 is a diagram for conceptually explaining details of the process of evaluating the reliability of the outline by the reliability evaluation unit. 4 is a flowchart illustrating the flow of a mounting operation for a vehicle according to the present embodiment. 10 is a flowchart illustrating the flow of a relative position detection process according to the present embodiment. 2 is a side view showing a state in which the front wheels of a vehicle are mounted on a mounting portion of the automatic transport device shown in FIG. 1.

The following describes an automatic transport device equipped with a vehicle position evaluation device according to one embodiment of the present invention, and the details of an automatic transport system equipped with this automatic transport device, with reference to the accompanying drawings.

Figure 1 shows the overall configuration of an automated vehicle transport system 10 according to one embodiment of the present invention. This automated transport system 10 transports vehicles C completed in a factory F to a container yard Y, which is a vehicle waiting area, and mainly comprises multiple automated transport devices 11 and a control unit 12 that controls the operation of the multiple automated transport devices 11. In this embodiment, the left and right front wheels of vehicle C, which is a front-wheel drive vehicle, are mounted on the automated transport device 11, and vehicle C is transported with its left and right rear wheels in contact with the ground. Details will be explained below as an example.

As shown in Figures 2 to 4, the automated guided vehicle 11 includes two drive wheels 13, a power applying unit 14 that applies power to each drive wheel 13, a mounting unit 15 capable of mounting one of the front wheels FW and rear wheels RW of the vehicle C (here, the left and right front wheels FW, which are the drive wheels), a satellite positioning system receiver 16, and a detector 17. In this embodiment, the detector 17 corresponds to the point cloud data acquisition unit of the present invention. Each drive wheel 13 is housed in a casing 18. These two casings 18 are connected to each other by a connecting unit 19. In this case, the mounting units 15 are provided at predetermined widthwise positions on the connecting unit 19, i.e., at two widthwise positions corresponding to the front wheels FW of the vehicle C to be mounted. Therefore, in this case, the mounting unit 15 is disposed between the two drive wheels 13. In other words, the mounting unit 15 is disposed in a position that overlaps with the two drive wheels 13 in the fore-and-aft direction of the vehicle body (see Figure 3). The width direction here refers to the direction perpendicular to the fore-and-aft direction of the vehicle body when the vehicle C is mounted, which corresponds to the direction in which the two drive wheels 13 of the automatic transport device 11 are separated from each other. Furthermore, the vehicle C here includes not only completed vehicles that are ultimately available for purchase by consumers, but also mass-produced vehicles that do not have the open cargo bed or box-shaped cargo compartment of a light truck, in other words, vehicles in an unmounted state (vehicles before mounting).

In this embodiment, each drive wheel 13 is composed of two wheels 20 arranged in parallel to each other, as shown in Figures 3 and 4. Each wheel 20 is rotatable independently of the other and is rotatably supported on a rotation shaft 21 disposed between the wheels 20. The rotation shaft 21 is disposed with its longitudinal direction aligned with the vertical direction when the automated transport device 11 is traveling. Therefore, for example, by rotating the two wheels 20 in opposite directions or by driving only one wheel 20, the drive wheel 13 having the two wheels 20 rotates around the rotation shaft 21, enabling steering of the automated transport device 11. Note that in this embodiment, the two wheels 20 constituting the drive wheels 13 are rotatable by 90 degrees or more (e.g., 360 degrees) around the rotation shaft 21. Therefore, for example, by rotating these two wheels 20 by 90 degrees (not shown), the direction of travel of the automated transport device 11 can be changed from the fore-and-aft direction of the vehicle to the width direction.

In this embodiment, as shown in Figures 3 and 4, the power applying unit 14 is provided in the same number as the wheels 20 and is composed of multiple in-wheel motors 22 that apply rotational driving force to each wheel 20. Each in-wheel motor 22 has, for example, a stator and a rotor (not shown), with the stator fixed to the rotation shaft 21 and the rotor connected to the wheel 20. As a result, by rotating the rotor of the in-wheel motor 22, the wheel 20 connected to the rotor can rotate together with the rotor. Furthermore, as described below, each in-wheel motor 22 can be electrically controlled independently. Therefore, by configuring the power applying unit 14 as described above, it is possible to independently rotate two wheels 20 arranged in parallel, and it is also possible for the drive wheels 13 to independently rotate about both vertical and horizontal axes.

In addition, in this embodiment, auxiliary wheels 23 are disposed on the front side of the vehicle body of the connecting portion 19. In this case, the auxiliary wheels 23 are capable of contacting the ground further forward of the vehicle body of the automatic transport device 11 than the respective drive wheels 13, and rotate (roll relative to the ground) passively in response to the rotational drive of the drive wheels 13 in a ground contact state. Alternatively, as described below, when the automatic transport device 11 travels with the front wheels FW of the vehicle C mounted on the mounting portion 15, the auxiliary wheels 23 rotate passively in response to the rotational drive of the drive wheels 13 in a ground contact state.

As previously mentioned, the receiver 16 is capable of receiving signals from multiple satellites (not shown) for a satellite positioning system. By performing a predetermined calculation based on the received signals, the position on Earth of the vehicle C (of the receiver 16) at the time of reception can be obtained. The calculation for obtaining the position may be performed by the receiver 16 or by the control unit 12 (specifically, the terminal control unit 24 or the integrated control unit 25, described below). The receiver 16 may be attached to any position relative to the automated guided vehicle 11. For example, to ensure smooth communication with satellites (positioning satellites) for a satellite positioning system such as GPS, it is preferable to attach the receiver 16 as high as possible on the automated guided vehicle 11. In this embodiment, the receiver 16 is disposed at the tip (top end) of the erected portion 26 erected from the top surface of the casing 18 (see Figure 2, etc.).

The detection unit 17 is provided at a predetermined position on the automated guided vehicle 11. As described above, this detection unit 17 corresponds to the point cloud data acquisition unit of the present invention and is used to accurately determine the position of the vehicle C to be transported. In this case, the detection unit 17 is configured to acquire point cloud data related to position information on the surface of the vehicle C. For this reason, a laser sensor such as LiDAR is suitable for the detection unit 17. Furthermore, as described below, when considering acquiring point cloud data over a wide area of the surface of the vehicle C, a laser sensor that can acquire three-dimensional point cloud data by irradiating or scanning a laser in three dimensions is particularly suitable. Note that the detection unit 17 is not only used to accurately determine the position of the vehicle C to be transported as described above, but may also be used to accurately determine spatial information around the automated guided vehicle 11 (such as the presence or absence of approaching objects). In this case, the detection unit 17 may be provided with equipment for accurately detecting spatial information around the automated guided vehicle 11, separate from equipment for detecting the position of the vehicle C.

In this embodiment, the detection unit 17 is used to detect the position of the vehicle C. Taking this into consideration, it is preferable that the detection unit 17 be disposed on both sides of the width of the automatic transport device 11 so as to be able to acquire point cloud data for the left and right side surfaces Cs and the front surface Cf of the vehicle C to be transported, as shown in FIG. 6. Furthermore, as will be described later, when attempting to acquire point cloud data for the side surfaces Cs of the vehicle C over as wide an area as possible in the fore-and-aft direction of the vehicle body, it is preferable to dispose each detection unit 17 at a corresponding height position (bottom) of the automatic transport device 11 so as to be able to detect the lower part of the body B of the vehicle C, as shown in FIG. 5, for example. Of course, the number and arrangement of the detection units 17 are arbitrary as long as point cloud data for the required areas of the surface of the vehicle C can be acquired.

The control unit 12 controls the two drive wheels 13 and the power applying unit 14, enabling the automatic guided vehicle 11 to travel automatically with steering. In this embodiment, the control unit 12 is composed of multiple terminal control units 24 provided in each automatic guided vehicle 11 and an overall control unit 25 that communicates with these multiple terminal control units 24. Here, the terminal control unit 24 controls the drive wheels 13 and the power applying unit 14 based on commands from the overall control unit 25, position information (position data) of the automatic guided vehicle 11 obtained by the satellite positioning system via the receiving unit 16, information regarding the target route that will be the target for automatic travel, and spatial information about the surroundings of the automatic guided vehicle 11 obtained by the detection unit 17, thereby enabling the automatic guided vehicle 11 to travel automatically with steering. In other words, the terminal control unit 24 can timely control the steering angle and speed (including acceleration, deceleration, and stopping) of the automatic guided vehicle 11 based on the commands and position information described above.

In the above case, from the viewpoint of facilitating control by the control unit 12, it is desirable that the power applying unit 14 be an electrically driven device such as a motor. Furthermore, if the automatic transport device 11 is provided with a mechanism that performs the operations necessary for loading the vehicle C and transporting it while loaded, it is desirable that this mechanism be an electrically driven mechanism for the same reasons as the power applying unit 14.

As shown in FIG. 7, the position detection device 27 for the object to be detected includes a point cloud data acquisition unit (detection unit 17 in this case), a voxel generation unit 28, a data number reduction unit 29, an outline detection unit 30, and a position detection unit 31. In this embodiment, the position detection device 27 further includes a reliability evaluation unit 32. The position detection device 27 configured as described above is incorporated into, for example, the terminal control unit 24.

The voxel generation unit 28 is configured to generate a plurality of voxels B (see FIG. 9) in a cubic lattice pattern based on point cloud data related to position information, in this case, the primary point cloud data D1 (see FIG. 8) acquired by the point cloud data acquisition unit. At this time, the voxels B are generated in a unit lattice space in which the primary point cloud data D1 exists in a manner that satisfies predetermined conditions. In addition, at this time, the size of the generated voxels B can, in principle, be set arbitrarily. For example, the size of the voxels B is set on the order of 10 ¹ mm so that 20 to 200 (more preferably 30 to 100) voxels B are generated across the entire width of the front surface Cf of the vehicle C.

The data number reduction unit 29 is configured to reduce the number of primary point cloud data D1 located within each voxel B generated by the voxel generation unit 28 (see Figures 9 and 10). In this case, the amount of data reduction in each voxel B can be arbitrary. For example, the number of primary point cloud data D1 (secondary point cloud data D2) within each voxel B after reduction can be reduced to one. In this case, the primary point cloud data D1 closest to the center of each voxel B may be retained (see Figure 10, described later). Of course, the position of the secondary point cloud data D2 can be set arbitrarily. For example, although not shown, if there is only one secondary point cloud data D2, the secondary point cloud data D2 may be set at the center of each voxel B. Alternatively, the secondary point cloud data D2 may be set at a position that corresponds to the center of gravity (average position coordinates) of the primary point cloud data D1 located within each voxel B before reduction by the data number reduction unit 29. The voxel generation unit 28 and data number reduction unit 29 can also be combined into a single calculation processing unit, an example of which is a voxel grid filter.

When secondary point cloud data D2 obtained by the data number reduction unit 29 is input to the outline detection unit 30, the outline detection unit 30 is configured to apply a Hough transform to the input secondary point cloud data D2 to detect the provisional outlines Ls1, Lf1 of the vehicle C as the object to be detected (see Figure 11). Furthermore, when primary point cloud data D1 located around the provisional outlines Ls1, Lf1 from the primary point cloud data D1 is input to the outline detection unit 30, the outline detection unit 30 is configured to apply a Hough transform to the input primary point cloud data D1 around the provisional outlines Ls1, Lf1 to detect the real outlines Ls2, Lf2 of the vehicle C (see Figure 12).

In this embodiment, the provisional outline Ls1, Lf1 and the actual outline Ls2, Lf2 are detected based on the primary point cloud data D1 and secondary point cloud data D2 of the side Cs and front Cf of the vehicle C. Therefore, the provisional side line Ls1 and the provisional front line Lf1 are detected as the provisional outline Ls1, Lf1 (see FIG. 11), and the actual side line Ls2 and the actual front line Lf2 are detected as the actual outline Ls2, Lf2 (see FIG. 12). Furthermore, in this embodiment, when detecting the actual side line Ls2 and the actual front line Lf2, a Hough transform is applied to the primary point cloud data D1 in voxel B (hereinafter referred to as first voxel B1) through which the provisional side line Ls1 and the provisional front line Lf1 pass, thereby enabling the actual side line Ls2 and the actual front line Lf2 of the vehicle C to be detected. Of course, the manner in which the actual side line Ls2 and the actual front line Lf2 are detected is not limited to the manner described above. To shorten the calculation time, the primary point cloud data D1 may be narrowed down in any manner as long as a Hough transform is performed on a portion of the primary point cloud data D1. For example, the actual side line Ls2 and actual front line Lf2 of vehicle C may be detected by performing a Hough transform on the primary point cloud data D1 in voxel B1, through which the provisional side line Ls1 and provisional front line Lf1 pass, as well as the primary point cloud data D1 in voxel B adjacent to voxel B1.

The position detection unit 31 is configured to detect the position of vehicle C based on the actual outlines Ls2 and Lf2 detected by the outline detection unit 30. In this embodiment, the position detection unit 31 is configured to detect the position of vehicle C based on the left and right actual sidelines Ls2 and the actual frontline Lf2 (see Figure 6). In principle, the position of vehicle C to be detected can be any position. For example, as shown in Figure 6, a point (midpoint M) on the actual frontline Lf2 that is equidistant from each intersection I between the actual frontline Lf2 and the left and right actual sidelines Ls2 may be set, and the position of this midpoint M may be detected as the position of vehicle C. Furthermore, because each intersection I is a point on the actual frontline Lf2 (because the positions of multiple predetermined points on the actual frontline Lf2 are known), it is also possible to detect the position of vehicle C along with its posture. Of course, the detection method shown in Figure 6 is merely an example, and any detection method for vehicle C's position may be used as long as it can be detected based on at least one of the actual sidelines Ls2 and the actual frontline Lf2.

The reliability evaluation unit 32 is configured to evaluate the reliability of the actual outlines Ls2 and Lf2. In this embodiment, the reliability evaluation unit 32 is configured to evaluate the reliability of the actual sidelines Ls2 and the actual frontlines Lf2 as the actual outlines Ls2 and Lf2. In this case, as shown in FIG. 13, for example, the reliability evaluation unit 32 acquires the proportion of second voxels B2 containing secondary point cloud data D2 for voxels B (hereinafter referred to as second voxels B2) through which the actual sidelines Ls2 and the actual frontlines Lf2 pass. The reliability evaluation unit 32 is then configured to evaluate the reliability of the actual sidelines Ls2 and the actual frontlines Lf2 as higher, respectively, as the acquired proportion of second voxels B2 increases. For example, if the proportion is α [%], the closer α/100 is to 1, the higher the reliability of the actual sidelines Ls2 and the actual frontlines Lf2 as higher.

Also, in this embodiment, as shown in FIG. 13, the reliability evaluation unit 32 is configured to obtain the proportion of third voxels B3 in which secondary point cloud data D2 exists for voxels B (hereinafter referred to as third voxels B3) adjacent to second voxels B2 on the side farther from the vehicle C that is the subject of detection (the left and bottom side in FIG. 13), and to evaluate the reliability of the main side line Ls2 and the main front line Lf2 as higher the lower the obtained proportion of third voxels B3. For example, if the above proportion is β [%], the closer 1-β/100 is to 1, the higher the reliability of the main side line Ls2 and the main front line Lf2 is evaluated as higher.

The reliability evaluation unit 32 may also be configured to acquire the proportion of fourth voxels B4 containing secondary point cloud data D2 for voxels B (hereinafter referred to as fourth voxels B4) adjacent to the second voxel B2 on the vehicle C side (the center side in Figure 13) among the multiple voxels B, and to evaluate that the higher the proportion of acquired fourth voxels B4, the higher the reliability of the main side line Ls2 and the main front line Lf2.

Next, an example of the operation of an automatic transport device 11 equipped with a position detection device 27 configured as described above, and an automatic transport system 10 for a vehicle C equipped with this automatic transport device 11, will be described primarily with reference to Figures 5 to 15.

First, the central control unit 25 sends commands to multiple automated guided vehicles 11 regarding what each automated guided vehicle 11 should do next. For example, in FIG. 1, the central control unit 25 sends a command to the automated guided vehicle 11 (11a) located between container yard Y and factory F, which is empty (i.e., does not have vehicle C loaded), to move toward factory F. The central control unit 25 also sends a command to the automated guided vehicle 11 (11b) located within the premises of factory F to move toward a position where the vehicle C to be transported can be loaded. The central control unit 25 also sends a command to the automated guided vehicle 11 (11c) located between factory F and container yard Y and loaded with vehicle C to move toward a predetermined waiting position within container yard Y, and to the automated guided vehicle 11 (11d, 11e) after unloading, to change its orientation relative to the direction of movement as necessary and then to retreat from container yard Y. Upon receiving the command, each automated guided vehicle 11 (11a-11e) controls its drive wheels 13 and power applying unit 14 via the terminal control unit 24 so that it operates in accordance with the command. In this way, vehicle C is continuously transported automatically by multiple automatic transport devices 11 (11a-11e).

Next, an example of the operation of loading vehicle C onto the automatic transport device 11 will be described in detail.

Figure 14 is a flowchart showing the workflow from when the automatic guided vehicle 11 approaches the vehicle C to be transported to when it loads the vehicle C, primarily illustrating the operation of the control unit 12 (here, the terminal control unit 24). According to this flowchart, the process of loading the vehicle C includes a relative position detection process S1 that detects the position of the vehicle C relative to the automatic guided vehicle 11, a movement process S2 that moves the automatic guided vehicle 11 based on the detected position of the vehicle C, and a loading process S3 that loads the vehicle C at a predetermined position on the automatic guided vehicle 11. The relative position detection process S1 here corresponds to the position detection method of the present invention. Below, we will explain the details of each process S1 to S3, focusing on the relative position detection process S1.

(S1) Relative Position Detection Step As shown in Fig. 15, the relative position detection step S1 includes a point cloud data acquisition step S11, a voxel generation step S12, a data number reduction step S13, a provisional outline detection step S14, a real outline detection step S15, and a position detection step S16. In this embodiment, the relative position detection step S1 further includes a reliability evaluation step S17. The above-mentioned relative position detection step S1 is performed using a position detection device 27 incorporated in the terminal control unit 24.

(S11) Point Cloud Data Acquisition Step In this step, the point cloud data acquisition unit acquires point cloud data related to position information on the surface of the vehicle C. In this embodiment, the point cloud data acquisition unit is configured with the detection unit 17. Therefore, as shown in FIGS. 5 and 6, for example, the detection unit 17 emits a laser L in three dimensions to detect point cloud data related to position coordinates on the surface of the vehicle C, including the side Cs and front Cf of the vehicle C. FIG. 8 shows an example of an image including the point cloud data (primary point cloud data D1) acquired at this time. This image shows the distribution of the primary point cloud data D1 acquired at the corner Cc of the vehicle C (see FIG. 6) and the side Cs and front Cf around the corner Cc. As such, the actual primary point cloud data D1 does not necessarily represent position coordinates on the surface of the vehicle C, but rather has a form distributed in the fore-and-aft direction, height direction, or width direction of the vehicle body.

(S12) Voxel Generation Step In this step, voxels B are generated based on the primary point cloud data D1 acquired in the point cloud data acquisition step S11. Specifically, the primary point cloud data D1 detected by the detection unit 17 is input to the voxel generation unit 28, which generates three-dimensionally distributed voxels B based on the input primary point cloud data D1. As a result, a plurality of voxels B are generated in the image including the primary point cloud data D1 according to the distribution pattern of the primary point cloud data D1 (see FIG. 9).

(S13) Data Number Reduction Step In this step, the number of pieces of primary point cloud data D1 located in each voxel B is reduced. Specifically, the number of pieces of primary point cloud data D1 in each voxel B in an image in which multiple voxels B have been generated by the data number reduction unit 29 is reduced (see FIG. 10 ). In this embodiment, regardless of the number of pieces of primary point cloud data D1 located in each voxel B, the number of pieces of primary point cloud data D1 located in all voxels B is reduced to one (see FIG. 10 ).

(S14) Provisional Outline Detection Step In this step, the outline detection unit 30 performs a Hough transform on the data-reduced primary point cloud data D1 (secondary point cloud data D2) obtained in the data number reduction step S13, and detects provisional outlines Ls1, Lf1 of the vehicle C, which are characteristic straight lines, from the image including the secondary point cloud data D2. Here, the primary point cloud data D1, which is the source of the secondary point cloud data D2, is obtained as position coordinate data of the side Cs and front Cf of the vehicle C, so that the provisional outlines Ls1, Lf1 detected are a provisional side line Ls1, which is a straight line along the side Cs, and a provisional front line Lf1, which is a straight line along the front Cf (see FIGS. 6 and 11 ).

(S15) Final Outline Detection Step In this step, the outline detection unit 30 applies a Hough transform to the primary point cloud data D1 located near the provisional side line Ls1 and the provisional front line Lf1 in the primary point cloud data D1, thereby detecting the final outline Ls2, Lf2 of the vehicle C, which is a characteristic straight line, from the image including the primary point cloud data D1. In this case, for example, as shown in FIG. 12, the Hough transform may be applied to the primary point cloud data D1 located within a first voxel B1 through which the provisional side line Ls1 and the provisional front line Lf1 pass. Also in the final detection step S15, the primary point cloud data D1 to be subjected to the Hough transform is acquired as position coordinate data of the side Cs and the front Cf of the vehicle C. Therefore, the final outline Ls2, Lf2 is detected as a straight line along the side Cs and a straight line along the front Cf (see FIG. 12).

In this case, from the perspective of improving the detection accuracy of the actual sideline Ls2 and the actual frontline Lf2, the primary point cloud data D1 located within voxel B (voxel B shown cross-hatched in Figure 12) where the intersection of the provisional sideline Ls1 and the provisional frontline Lf1 is located may be removed, and then the primary point cloud data D1 after removal may be subjected to a Hough transform. Furthermore, the range of the primary point cloud data D1 that is the subject of the Hough transform may be limited to the height direction region around the provisional outlines Ls1 and Lf1.

(S17) Reliability Evaluation Step In this step, the reliability evaluation unit 32 evaluates the reliability of the final sideline Ls2 and the final frontline Lf2 detected in the final outline detection step S15. In this embodiment, the voxel generation unit 28 first generates multiple voxels B based on the primary point cloud data D1. At this time, the primary point cloud data D1 used to generate the voxels B may be limited to a range around the final outlines Ls2 and Lf2 (e.g., a range in the width direction and the front-rear direction of the vehicle body) compared to that used in the voxel generation step S12. Furthermore, due to the nature of the reliability evaluation step S17, if it is considered acceptable to lower the density of the primary point cloud data D1 used compared to that used when detecting the provisional outlines Ls1 and Lf1, the size of the generated voxels B may be larger than that used in the voxel generation step S12. Then, the data number reduction unit 29 reduces the number of primary point cloud data D1 located within each generated voxel B to obtain secondary point cloud data D2. In this case, as in the data number reduction step S13, the number of data items in the primary point cloud data D1 is reduced so that the number of secondary point cloud data items D2 in each voxel B becomes one.

After generating the multiple voxels B and secondary point cloud data D2 as described above, the reliability evaluation unit 32 evaluates the reliability of the actual sideline Ls2 and the actual frontline Lf2. Specifically, as shown in FIG. 13, for each of the multiple voxels B, where the secondary point cloud data D2 exists, a percentage α [%] of the multiple voxels B is obtained for each second voxel B2 through which the actual sideline Ls2 and the actual frontline Lf2 pass. In this embodiment, for each of the multiple voxels B, where the secondary point cloud data D2 exists, a percentage β [%] of the multiple voxels B is obtained for each second voxel B3 adjacent to the second voxel B2 on the side farther from the vehicle C that is the subject of detection. The closer α/100 × (1 - β/100) is to 1, the higher the reliability of the actual sideline Ls2 and the actual frontline Lf2 is evaluated to be.

In the reliability evaluation step S17, although not shown in the figure, the proportion of voxels for which secondary point cloud data D2 exists may be obtained for voxels B among the multiple voxels B generated by the voxel generation unit 28 that are at different positions in the height direction of the vehicle C from the main side line Ls2 and the main front line Lf2, and based on the obtained proportion, an evaluation may be made as to whether the subject from which the primary point cloud data D1 was obtained is vehicle C (the reliability of vehicle C).

(S16) Position Detection Step If the reliability of the actual sideline Ls2 and the actual front line Lf2 is evaluated to be high (e.g., a value equal to or greater than a predetermined threshold) in the reliability evaluation step S17, the position of the vehicle C is detected based on these actual sideline Ls2 and the actual front line Lf2. In this embodiment, the position of a point (midpoint M) on the actual front line Lf2 that is equidistant from each intersection I of the actual front line Lf2 and the left and right actual side lines Ls2 is detected as the position of the vehicle C (see FIG. 6). Furthermore, if the positions and inclinations of the actual sideline Ls2 and the actual front line Lf2 have been acquired, the position together with the attitude of the vehicle C (the relative position of the vehicle C with respect to the automatic guided vehicle 11) is detected from this information.

In addition, in this embodiment, the accurate position of the automatic guided vehicle 11 is acquired by a satellite positioning system, so the position (absolute position) of vehicle C can be detected from the position (absolute position) of this automatic guided vehicle 11 and the position (relative position) of vehicle C acquired by the position detection unit 31.

(S2) Movement step After detecting the position of the vehicle C as described above, the automatic transport device 11 is moved based on the detected position of the vehicle C. Specifically, assuming that the detected vehicle C will be loaded onto the loading unit 15, the automatic transport device 11 is moved toward the vehicle C in a predetermined orientation.

(S4) Loading Possibility Determination Step Then, when the automatic transporting device 11 has moved a predetermined distance, it determines whether or not the automatic transporting device 11 has moved to a position where the vehicle C can be loaded. If it determines that the automatic transporting device 11 has not yet moved to a position where the vehicle C can be loaded, the relative position detection step S1 and the movement step S2 are performed again. In this way, the relative position detection step S1 and the movement step S2 are repeatedly performed at predetermined time intervals, and if it is determined in the loading possibility determination step S4 that the automatic transporting device 11 has moved to a position where the vehicle C can be loaded, the process proceeds to the step of loading the vehicle C onto the automatic transporting device 11 (loading step S3).

(S3) Vehicle Mounting Step In this step, the automatic guided vehicle 11 and the vehicle C are brought closer to each other based on the position of the vehicle C obtained in the relative position detection step S1, thereby mounting the front wheels FW of the vehicle C on the mounting unit 15 (see FIG. 16 ). This allows the vehicle C to be mounted at a position where it can be automatically transported by the automatic guided vehicle 11. The above-described mounting operation may be performed by the automatic guided vehicle 11 approaching a stopped vehicle C, or by the vehicle C approaching the stopped automatic guided vehicle 11. Alternatively, the above-described mounting operation may be performed by a moving body different from either the automatic guided vehicle 11 or the vehicle C entering underneath the vehicle C and lifting (tilting) a portion of the vehicle C using a lift-up mechanism provided on the moving body, and then the automatic guided vehicle 11 approaching the vehicle C. Alternatively, the above-described mounting operation may be performed by moving the vehicle C above a lift-up mechanism attached to a predetermined position, and then lifting a portion of the vehicle C using the lift-up mechanism, and then the automatic guided vehicle 11 approaching the vehicle C. In addition, as in this embodiment, when the detection direction of the detection unit 17 is set toward the forward movement direction of the automatic conveying device 11 (see Figures 5 and 6), the automatic conveying device 11 can be brought close to the vehicle C, stopped at a position where the vehicle C can be mounted, and then the automatic conveying device 11 can be rotated around the vertical axis at the stopped position to flip the front and back, making the vehicle C ready to be mounted on the mounting unit 15.

As described above, the position detection device 27 and position detection method according to this embodiment generate multiple voxels B based on the acquired primary point cloud data D1, reduce the number of primary point cloud data D1 located within each voxel B, and perform a Hough transform on the reduced primary point cloud data D1, i.e., secondary point cloud data D2, to detect the provisional outlines Ls1 and Lf1 of the vehicle C as the object to be detected. This reduces the number of primary point cloud data D1 used in the Hough transform, enabling detection of the provisional outlines Ls1 and Lf1 to be completed in a short time. Meanwhile, the position of the vehicle C is detected by detecting the actual outlines Ls2 and Lf2 using a Hough transform on the original primary point cloud data D1, thereby enabling highly accurate position detection. Furthermore, by limiting the primary point cloud data D1 used for position detection to the area surrounding the provisional outlines Ls1 and Lf1, only the primary point cloud data D1 likely to be relevant to the detection of the actual outlines Ls2 and Lf2 can be used from the original primary point cloud data D1. Therefore, the position of vehicle C can be detected efficiently and in a short time while achieving high position detection accuracy, which makes it possible to load vehicle C onto the automatic transport device 11 accurately and quickly.

Furthermore, if the object to be detected is a vehicle C, as in this embodiment, the Hough transform of the outline detection unit 30 can accurately detect the outline along the side Cs and front Cf of the vehicle C, i.e., the side lines Ls1, Ls2 and front lines Lf1, Lf2, with a very high probability. Therefore, by applying the position detection device 27 or position detection method according to this embodiment to detecting the position of the vehicle C, it is possible to accurately detect the position of the vehicle C, along with its attitude.

Furthermore, when the object to be detected is a vehicle C as described above, in this outline detection process S15, the primary point cloud data D1 located within voxel B (voxel B shown cross-hatched in Figure 12) where the intersection of the provisional side line Ls1 and provisional front line Lf1 is located is removed, in other words, the primary point cloud data D1 on the surface of the corner portion Cc of the vehicle C. After that, a Hough transform is applied to the primary point cloud data D1 after removal, thereby further improving the position detection accuracy of the actual side line Ls2 and the actual front line Lf2.

In addition, the automated guided vehicle 11 according to this embodiment further includes a reliability evaluation process S17 for evaluating the reliability of the actual sideline Ls2 and the actual frontline Lf2. This increases the reliability of the actual sideline Ls2 and the actual frontline Lf2 detected in the outline detection process S15. This also increases the reliability of the position of vehicle C detected based on the actual sideline Ls2 and the actual frontline Lf2. Furthermore, for at least the voxels B (second voxels B2) through which the actual sideline Ls2 and the actual frontline Lf2 pass, a proportion α of the second voxels B2 containing secondary point cloud data D2 is obtained. By evaluating the reliability of the actual sideline Ls2 and the actual frontline Lf2 based on this proportion α, the reliability of the actual sideline Ls2 and the actual frontline Lf2 can be evaluated efficiently and quickly in process S17.

The above describes one embodiment of the present invention, but the position detection method and position detection device according to the present invention may also have configurations other than those described above, as long as they do not deviate from the spirit of the invention.

For example, in the above embodiment, a reliability evaluation process S17 is provided after the main outline detection process S15, and the position of vehicle C is detected after evaluating the reliability of the main side line Ls2 and main front line Lf2 detected in the main outline detection process S15. However, depending on the conditions, the reliability evaluation process S17 may be omitted, and a position detection process S16 may be provided immediately after the main outline detection process S15.

In addition, while the above embodiment illustrates an example in which the control unit 12 is configured with the terminal control unit 24 and the overall control unit 25, this is not necessarily limited to this. For example, the overall control unit 25 may be omitted, and the automatic travel of the automatic transport device 11 may be controlled solely by the terminal control unit 24. In this case, for example, for the automatic transport device 11, a program related to automatic transport involving loading and unloading may be stored in advance in the terminal control unit 24, and the drive wheels 13 and power applying unit 14 may be driven and controlled based on the program through predetermined operations by an operator.

Furthermore, the automated guided transport system 10 shown in Figure 1 illustrates a case in which only the front wheels FW are mounted on the mounting section 15 as part of the wheels of the vehicle C, but this is of course not limited to this. For example, although not shown, the present invention may be applied when only the rear wheels RW of the vehicle C are mounted on the mounting section 15. Alternatively, the present invention may be applied when both the front wheels FW and rear wheels RW of the vehicle C are mounted on the mounting section 15.

In addition, in the above embodiment, a position detection device 27 according to the present invention is provided on an automatic transport device 11 for automatically transporting a vehicle C, and the automatic transport device 11 approaches the vehicle C while detecting the position of the vehicle C using the position detection device 27. However, this is not necessarily limited to this. For example, the present invention may be applied when a vehicle C approaches a stopped automatic transport device 11. Alternatively, the present invention may be applied when a moving body other than the automatic transport device 11 approaches the vehicle C, such as when a moving body is brought close to the vehicle C in a factory to perform a predetermined task on the vehicle C, and the position detection device 27 according to the present invention is provided on the moving body. Of course, the present invention may also be applied when an object other than the vehicle C is the subject of position detection (detectable body), as long as it has an external shape whose outline can be detected using a Hough transform.

10 Automatic transport system 11 Automatic transport device 12 Control unit 13 Drive wheel 14 Power supply unit 15 Mounting unit 16 Receiving unit 17 Detection unit 18 Casing 19 Connecting unit 20 Wheel 21 Rotating shaft 22 In-wheel motor 23 Auxiliary wheel 24 Terminal control unit 25 Overall control unit 26 Standing unit 27 Position detection device 28 Voxel generation unit 29 Data number reduction unit 30 Outline detection unit 31 Position detection unit 32 Reliability evaluation unit B, B1, B2, B3, B4 Voxel C Vehicle Cc Corner portion Cf Side Cs Side D1 Primary point cloud data D2 Secondary point cloud data F Factory FW Front wheel L Laser Lf1 Provisional front line Lf2 Actual front line Ls1 Provisional side line Ls2 Actual side line S1 Relative position detection process S11 Point cloud data acquisition process S12 Voxel generation process S13 Data number reduction process S14, temporary outline detection process S15, real outline detection process S16, position detection process S17, reliability evaluation process S2, movement process S3, loading process S4, loading possibility determination process Y, container yard

Claims (3)

a point cloud data acquisition step of acquiring primary point cloud data relating to position information of the surface of a vehicle as a detected object by a point cloud data acquisition unit;
a voxel generation step of generating a plurality of voxels based on the primary point cloud data by a voxel generation unit ;
a data number reduction step of reducing the number of the primary point cloud data located in each of the voxels by a data number reduction unit;
a tentative outline detection step of detecting a tentative outline of the object to be detected by applying a Hough transform to the secondary point cloud data obtained in the data number reduction step by an outline detection unit ;
a main outline detection step of detecting a main outline of the object by applying a Hough transform to the primary point cloud data located around the provisional outline among the primary point cloud data by the outline detection unit;
a position detection step of detecting the position of the object to be detected based on the outline by a position detection unit ,
In the provisional outline detection step, a linear provisional side line is detected as the provisional outline based on the secondary point cloud data of the side of the vehicle, and a linear provisional front line that is perpendicular to the provisional side line is detected based on the secondary point cloud data of the front of the vehicle,
A vehicle position detection method in which, in the main outline detection process, a linear main side line is detected as the main outline based on primary point cloud data of the side of the vehicle, and a linear main front line that is perpendicular to the main side line is detected based on primary point cloud data of the front of the vehicle . a reliability evaluation step of evaluating the reliability of the outline by a reliability evaluation unit ;
The position detection method described in claim 1, wherein the reliability evaluation process obtains the proportion of voxels in which the secondary point cloud data exists for at least the voxels through which the outline passes among the plurality of voxels, and evaluates that the reliability of the outline is higher the higher the proportion. a point cloud data acquisition unit capable of acquiring primary point cloud data relating to position information of the surface of a vehicle as a detected object;
a voxel generation unit that generates a plurality of voxels based on the primary point cloud data;
a data number reduction unit that reduces the number of the primary point cloud data located in each of the voxels;
an outline detection unit that applies a Hough transform to the secondary point cloud data obtained by the data number reduction unit to detect a provisional outline of the object to be detected, and applies the Hough transform to primary point cloud data located around the provisional outline among the primary point cloud data to detect a real outline of the object to be detected;
a position detection unit capable of detecting the position of the object to be detected based on the outline ,
The outline detection unit is a vehicle position detection device configured to detect, as the temporary outline, a linear temporary side line based on secondary point cloud data of the side of the vehicle, and a linear temporary front line that is perpendicular to the temporary side line based on secondary point cloud data of the front of the vehicle, and to detect, as the real outline, a linear actual side line based on primary point cloud data of the side of the vehicle, and a linear actual front line that is perpendicular to the real side line based on primary point cloud data of the front of the vehicle .