JP7664625B2 - DETECTION AREA SETTING DEVICE, OBSTACLE DETECTION SYSTEM, AND DETECTION AREA SETTING METHOD - Google Patents

Description

The present invention relates to a detection area setting device, an obstacle detection system, and a detection area setting method.

Patent Document 1 discloses a movable home gate device that aims to improve the safety of train operations by updating the detection area for obstacle detection by the detector of the movable home gate device for each train that stops at the platform.

The movable home gate device is a movable home gate device in which a fence is provided along the edge of the track at the platform, and the portion facing the side door at the train's stopping position is an opening that can be opened and closed, and includes a detection device provided for each opening, the detection area of which includes the area between the outside of the fence and the train stopped at the platform, and a control means for the detection device, and the control means is characterized in that it has a detection area update function that updates the detection area of the detection device in response to the position and shape of the side of the vehicle for each train stopped at the platform or for each vehicle that constitutes the train formation.

That is, the control means measures the distance from the irradiation starting point of the detection device to each part of the side of the train, and identifies the shape and position of the side of the train from the measured distance to each part of the side of the train. Next, the control means determines the detection area for detecting obstacles for the train currently parked at the platform to be a range that corresponds to the identified shape and position of the side of the train. Specifically, the control means determines the detection area so that a surface having a shape that conforms to the side of the train, within the range included in the side shape detection range, is formed on the platform side in the train width direction, at a position a predetermined distance away from the side of the train, and updates it as the detection area for detecting obstacles for the detection device.

The detection device used is LiDAR (Light Detection and Ranging), which measures the distance to an object by scanning a laser beam at the object and measuring the scattered and reflected light.

JP 2013-203118 A

The detection area updating method disclosed in Patent Document 1 updates the pre-set side shape detection range to a new side shape detection range that is a predetermined distance away from the detected vehicle side shape, which has the following inherent problems:

To update the detection area, the control means had to determine a shape that was a predetermined distance away from the shape of the vehicle side detected by the detection device. To do this, the control means had to execute very complicated and highly loady arithmetic processing to calculate a shape that was a uniform predetermined distance away from the vehicle side, based on the measured distances to each part of the vehicle side.

In addition, the side shape detection range had to be set in advance, and the task of setting it individually based on the relative positions of the detection device and the object to be detected was extremely cumbersome and time-consuming.

Furthermore, when the control means detects the shape of the vehicle's side surface via the detection device, the laser light emitted from the detection device may pass through the window glass in the vehicle's door, and reflected light from other objects may be detected through the window glass, which creates an inherent problem in that the control means is unable to accurately detect the shape of the vehicle's side surface.

A similar problem occurs not only in the detection device of a movable home gate device, but also in various other situations where the reference area for setting the detection area changes. For example, the detection area is set when a detection device detects obstacles in the vicinity of a vehicle parked on a lift in a multi-story parking garage, and the vehicle is moved to a storage facility if no obstacles are present. The detection area must be updated depending on the vehicle type and parking position, and similar problems exist.

In view of the above-mentioned conventional technology, the object of the present invention is to provide a detection area setting device, an obstacle detection system, and a detection area setting method that can accurately and easily update a detection area even when the reference object to be measured for which the detection area is set changes.

In order to achieve the above-mentioned object, the first characteristic configuration of the detection area setting device according to the present invention is configured to include a point cloud data acquisition unit that acquires three-dimensional point cloud data indicating the coordinates of distance measurement points including reflection points on the surface of a reference object based on the reflected light of the measurement light irradiated on an area including a reference object to be set as a detection area, a surface acquisition unit that acquires multiple surfaces including the surface of the reference object based on the point cloud data acquired by the point cloud data acquisition unit, a labeling unit that labels the multiple surfaces acquired by the surface acquisition unit under predetermined conditions, and an area setting unit that sets the detection area for detecting obstacles based on the multiple surfaces labeled by the labeling unit, and the surface acquisition unit executes a process for acquiring, for multiple attention points, a normal to an attention surface generated at an attention point and a point near the attention point among the multiple distance measurement points included in the point cloud data as a normal vector at the attention point, and executes a surface acquisition process that acquires multiple surfaces including any of the attention points based on each normal vector.

Based on the three-dimensional point cloud data indicating the coordinates of the distance measurement points relative to the area including the reference object acquired by the point cloud data acquisition unit, a surface acquisition unit acquires multiple surfaces including the surface of the reference object. A detection area for detecting obstacles is set by the area setting unit based on the multiple surfaces labeled by the labeling unit.

In other words, the detection area can be defined based on multiple labeled surfaces rather than distance data from individual ranging points, so the detection area can be updated accurately and easily even if the shape or position of the reference object contained in the reference area used to set the detection area changes.

Then, the surface acquisition unit executes a surface acquisition process. In the surface acquisition process, a surface of interest is generated at a point of interest and multiple points adjacent to the point of interest among the multiple distance measurement points included in the point cloud data, and a normal to the surface of interest is acquired as a normal vector at the point of interest. In the surface acquisition process, a similar process is repeated for multiple points of interest. This makes it easy to acquire multiple surfaces that include any of the points of interest based on each normal vector.

The second characteristic configuration is that, in addition to the first characteristic configuration described above, the surface acquisition process includes a clustering process that groups other points of interest that exist within a predetermined search range from a specific point of interest in the same group as the specific point of interest if the normal vectors of the other points of interest fall within a predetermined allowable range of inclination with respect to the direction of the normal vector of the specific point of interest.

In the clustering process, other points of interest that exist within a specified search range are extracted based on a specific point of interest, and when the inclination of the normal vector of the other points of interest relative to the normal vector of the specific point of interest falls within a specified tolerance range, the other points of interest are grouped into the same group as the specific point of interest as candidate points that constitute the same surface as the surface that contains the specific point of interest. The search range is a value that is set appropriately depending on the reference object to be measured. For example, if the reference object to be measured is the side of a train, the search range is set to a value that excludes points of interest in areas far away from the side of the train, while causing multiple points of interest corresponding to both sides of the train's window area to belong to the same group.

The third characteristic configuration is that, in addition to the second characteristic configuration described above, the surface acquisition process repeats the clustering process with each attention point as the specific attention point, and if any of the attention points to be grouped in a later clustering process have already been grouped in a previous clustering process, the surface acquisition process includes an integration process that integrates all attention points to be grouped in the later clustering process into the same group as the previous clustering process.

When the clustering process is repeated with each focus point as a specific focus point, a cluster is formed for each specific focus point that includes other focus points that exist on the same surface. If any of the focus points that make up each cluster belong to a previously formed cluster, they are merged into the same cluster as the previously formed cluster, thereby increasing the number of focus points that are candidates for being included in the same surface. In other words, the surface generated in the specified search range is expanded beyond the specified search range.

The fourth characteristic configuration is that, in addition to the second or third characteristic configuration described above, the surface acquisition process includes a consolidation process in which a representative normal vector representing each of the multiple groups generated by the clustering process is obtained, and groups whose respective representative normal vectors fall within a predetermined tolerance range are consolidated into the same group.

The attention points grouped by the second or third characteristic configuration are a collection of multiple attention points that exist within a specified search range based on at least a specific attention point. If the search range is set large, attention points that actually belong to different surfaces may belong to the same group, so the search range must be limited to a certain size. However, if the search range is set small, attention points that actually belong to the same surface may belong to different groups. Therefore, in this configuration, if the representative normal vectors of each group fall within a specified allowable range, they are configured to be merged into the same group. The specified allowable range can be specified, for example, by the orientation of the representative normal vector, the distance between the starting points of the representative normal vectors relative to a specified reference surface, etc.

The fifth characteristic configuration is that, in addition to any one of the second to fourth characteristic configurations described above, the surface acquisition process performs a filtering process on the points of interest grouped into the same group to remove outliers from the same group.

Point cloud data grouped into the same group by the clustering process is a collection of attention points whose distance between each ranging point and the inclination of the normal direction are within a specified tolerance range, and the extent to which each attention point is away from the normal direction of a specific attention point or the tangent plane of a specific attention point is not evaluated. Therefore, by performing a filtering process, outliers that cannot be evaluated as existing on the same plane are appropriately removed.

The sixth characteristic configuration is that, in addition to the fifth characteristic configuration described above, the filtering process is a process that performs robust estimation including any of RANSAC, M-estimation, or the least median method.

Robust estimation can be preferably used as this type of filtering process. Robust estimation is a statistical processing method that aims to evaluate the possibility that a given observation value contains an outlier and suppress its influence. Specifically, any of the RANSAC, M-estimation, or least median method can be preferably used.

The seventh characteristic feature of the present invention is that, in addition to any one of the first to sixth characteristic features described above, the surface acquisition process acquires a surface based on points of interest included in the same group.

The eighth characteristic configuration of the present invention is that, in addition to any one of the first to seventh characteristic configurations described above, the labeling unit identifies and labels a reference plane corresponding to the surface of the reference object from among the multiple surfaces and an intersecting plane that intersects with the reference plane based on a preset reference vector and the normal vectors of the multiple surfaces acquired by the surface acquisition unit.

The labeling section identifies the surfaces whose normal vectors have a predetermined relationship with respect to a preset reference vector, specifically the reference surface corresponding to the surface of the reference object to be measured, and the intersecting surface that intersects with the reference surface, and labels each of these as two surfaces that define the detection area.

The ninth characteristic configuration is that, in addition to any one of the first to seventh characteristic configurations described above, the labeling unit identifies a reference surface corresponding to the surface of the reference object from the multiple surfaces based on a preset reference vector and the normal vectors of the multiple surfaces acquired by the surface acquisition unit, acquires a boundary line of the reference surface based on the point cloud data acquired by the point cloud data acquisition unit, generates an intersecting surface that passes through the boundary line and intersects with the reference surface, and identifies and labels the reference surface and the intersecting surface that intersects with the reference surface.

The labeling unit identifies a surface whose normal vector has a predetermined relationship with a preset reference vector, specifically a reference surface that corresponds to the surface of a reference object. Furthermore, the labeling unit obtains the boundary line of the reference surface based on point cloud data including the reference surface, generates an intersection plane that passes through the boundary line and intersects with the reference surface, and labels the reference surface and the intersection plane that intersects with the reference surface so that they can be identified as two surfaces that define a detection area.

The tenth characteristic configuration is, in addition to the eighth or ninth characteristic configuration described above, that the area setting unit defines a detection area consisting of a polyhedron between the reference surface and the irradiation source of the measurement light based on the reference surface and the intersection surface labeled by the labeling unit.

Based on the labeled reference plane and intersection plane, a polyhedron that serves as the detection area is defined on the side of the measurement light source.

The first characteristic configuration of the obstacle detection system according to the present invention is that it includes a detection area setting device having any one of the first to tenth characteristic configurations described above, and a range sensor that scans a measurement light in an area including the detection area set by the detection area setting device and determines the presence or absence of an obstacle based on the presence or absence of reflected light from the detection area.

The range sensor determines whether or not an obstacle is present in the detection area defined by the detection area setting device.

The first characteristic configuration of the detection area setting method according to the present invention is configured to include a point cloud data acquisition step of acquiring three-dimensional point cloud data indicating the coordinates of distance measurement points including reflection points on the surface of a reference object based on the reflected light of the measurement light irradiated on an area including a reference object to be set as a detection area, a surface acquisition step of acquiring multiple surfaces including the surface of the reference object based on the point cloud data acquired in the point cloud data acquisition step, a labeling step of labeling the multiple surfaces acquired in the surface acquisition step under predetermined conditions, and an area setting step of setting the detection area for detecting obstacles based on the multiple surfaces labeled in the labeling step, in which the surface acquisition step executes a process for acquiring, for multiple attention points, a normal to an attention surface generated at an attention point and a point adjacent to the attention point among the multiple distance measurement points included in the point cloud data as a normal vector at the attention point, and acquires multiple surfaces formed of any of the attention points based on each normal vector.

The second characteristic configuration is, in addition to the first characteristic configuration described above, that the surface acquisition step includes executing a clustering process in which, if the normal vectors of other points of interest that exist within a predetermined search range from the specific point of interest are inclined within a predetermined allowable range with respect to the direction of the normal vector of the specific point of interest, the other points of interest are grouped together with the specific point of interest.

The third characteristic configuration is that, in addition to the second characteristic configuration described above, the surface acquisition step includes performing an integration process in which the clustering process is repeated with each attention point as the specific attention point, and if any of the attention points to be grouped in the subsequent clustering process have already been grouped in the previous clustering process, all attention points to be grouped in the subsequent clustering process are integrated into the same group as the previous clustering process.

The fourth characteristic configuration is that, in addition to the second or third characteristic configuration described above, the surface acquisition step includes performing a consolidation process in which a representative normal vector representing each of the multiple groups generated by the clustering process is obtained, and groups whose respective representative normal vectors have orientations within the predetermined tolerance range are consolidated into the same group.

The fifth characteristic configuration is, in addition to any one of the second to fourth characteristic configurations described above, that the surface acquisition step performs a filtering process on the points of interest that have been grouped into the same group in the clustering process to remove outliers from the same group.

The sixth characteristic configuration is that, in addition to the fifth characteristic configuration described above, the filtering process is a process that performs robust estimation including any of RANSAC, M-estimation, or the least median method.

The seventh characteristic configuration is that, in addition to any one of the first to sixth characteristic configurations described above, the labeling step identifies and labels a reference plane corresponding to the surface of the reference object from among the multiple surfaces and an intersecting plane that intersects the reference plane based on a preset reference vector and the normal vectors of the multiple surfaces acquired in the surface acquisition step.

The eighth characteristic configuration is that, in addition to any one of the first to sixth characteristic configurations described above, the labeling step identifies a reference surface corresponding to the surface of the reference object from the multiple surfaces based on a preset reference vector and the normal vectors of the multiple surfaces acquired in the surface acquisition step, acquires a boundary line of the reference surface based on the point cloud data acquired in the point cloud data acquisition step, generates an intersecting surface that passes through the boundary line and intersects with the reference surface, and identifies and labels the reference surface and the intersecting surface that intersects with the reference surface.

The ninth characteristic configuration is, in addition to the seventh or eighth characteristic configuration described above, that the area setting step defines a detection area consisting of a polyhedron between the reference surface and the irradiation source of the measurement light based on the reference surface and the intersection surface labeled in the labeling step.

As described above, the present invention makes it possible to provide a detection area setting device, an obstacle detection system, and a detection area setting method that can accurately and easily update a detection area even when the reference object to be measured for which the detection area is set changes.

(a) is an explanatory diagram of a side view of the obstacle detection area partitioned between the train and the platform safety fence, (b) is an explanatory diagram of a plan view of the same detection area, and (c) is an explanatory diagram overlooking the same detection area. 1A is an explanatory diagram of the functional blocks of the obstacle detection system, and FIG. 1B is an explanatory diagram of the LIDAR. 1A is an explanatory diagram of a detection area setting device (setting unit), and FIG. 1B is a flowchart showing a procedure for detection area setting processing. 13 is a flowchart showing the procedure of a surface acquisition process. 13 is a flowchart showing a procedure of a labeling process. 13 is a flowchart showing a procedure for a detection area setting process. 13 is a flowchart showing a procedure for an obstacle detection process. 4A, 4B, and 4C are explanatory diagrams of a surface acquisition process. FIG. 11 is an explanatory diagram of a first step of the detection area setting process. FIG. 11 is an explanatory diagram of a second step of the detection area setting process. FIG. 11 is an explanatory diagram of a third step of the detection area setting process. 13 is a flowchart showing the procedures of a surface acquisition process, a labeling process, and a general-purpose detection area setting process. 5A and 5B are explanatory diagrams of a surface acquisition process for setting a general-purpose detection area. FIG. 11 is an explanatory diagram of a platform edge recognition process. 5A, 5B, and 5C are diagrams illustrating a general-purpose detection area setting process.

Below, an example of a detection area setting device, an obstacle detection system, and a detection area setting method according to the present invention will be described.

[Detection area set on train platform]
As shown in Fig. 1(a), (b), and (c), a safety fence SF is installed on the edge of a platform P installed in a railway station building on the track side to prevent accidents such as a user coming into contact with a train. The safety fence SF is composed of a fence body SB with a gate provided at a position corresponding to the door 2 of the vehicle 1, and a door DR for opening and closing the gate. The X, Y, and Z coordinate systems shown in each figure indicate three-dimensional coordinate systems with the installation position of any one of the riders 10 as the origin. Fig. 1(a) shows the positional relationship between the vehicle 1, the platform P, and the safety fence SF as viewed along the Y-axis direction. Fig. 1(b) shows the positional relationship between the vehicle 1, the platform P, and the safety fence SF as viewed from above along the Z-axis direction.

A LiDAR (Light Detection and Ranging) 10 capable of two-dimensional scanning is installed above the door DR of the fence body SB so as to overlook the space around the door 2 of the vehicle 1.

As shown in FIG. 2(b), the LIDAR 10 employs a TOF system and includes a cylindrical casing 11 with an optical window W, a light projecting and receiving unit 13 housed in the casing 11 and equipped with a laser light source and a light receiving element made of a semiconductor laser, a scanning optical system 14, and a signal processing circuit 12 that controls them.

The scanning optical system 14 includes a deflection mirror M. The scanning optical system 14 also includes a first deflection mechanism 14A that oscillates the deflection mirror M around a first axis P1 that is perpendicular to the paper surface of FIG. 2(b). The first deflection mechanism 14A includes an electromagnetic oscillation mechanism. The scanning optical system 14 also includes a motor that rotates the second deflection mechanism 14B around a second axis P2 that is parallel to the paper surface of FIG. 2(b).

The scanning optical system 14 repeatedly scans the horizontal scanning line hd in the vertical direction vd via the first deflection mechanism 14A while repeatedly scanning the pulsed measurement light emitted from the laser light source in the horizontal direction hd within a predetermined range via the second deflection mechanism 14B. The signal processing circuit calculates the distance and direction to the reflection point of the measurement light based on the time difference between the emission timing of the measurement light and the detection timing of the reflected light. In other words, the LIDAR 10 detects the presence or absence of an obstacle based on the reflected light from an object present in the space surrounding the door 2 of the vehicle 1.

In the present invention, an obstacle detection area DA is set in the space scanned by the measurement light so as not to erroneously determine that the platform P or the vehicle 1 is an obstacle based on the light reflected from the surface of the platform P or the vehicle 1. Then, in the obstacle detection process in the obstacle detection system, which will be described in detail later using the flowchart of Figure 7, an object present within the detection area DA is detected as an obstacle based only on the light reflected from the object. As a result, passengers 3 getting on and off the vehicle 1 and baggage carried by the passengers 3 are detected as obstacles. If no obstacle is detected, the door 2 of the vehicle 1 and the gate door DR are safely closed.

The detection area DA is a region partitioned into a roughly rectangular parallelepiped shape with a predetermined width W set in the left-right direction along the above-mentioned side of the vehicle 1 based on the position of the rider 10, at a predetermined depth D set inside the platform P, based on an offset surface that is a predetermined width ΔD away from the side closest to the platform P, at a predetermined height ΔH from the floor surface of the platform P, and along the direction of travel of the vehicle 1. The detection area DA is an individual detection area that is set each time the vehicle 1 enters the platform P, because the position of the side of the vehicle facing the platform P differs depending on the model of the vehicle 1, etc.

[Configuration of Obstacle Detection System]
As shown in FIG. 2(a), the obstacle detection system 100 includes a detection area setting unit 20 that sets a detection area in which the presence or absence of an obstacle is determined, a range sensor that scans the detection area set by the detection area setting unit 20 with measurement light and determines the presence or absence of an obstacle based on the presence or absence of reflected light from the detection area, and a system control unit 50 that controls the operation timing of the detection area setting unit 20 and the range sensor.

The range sensor is made up of the lidar 10, which is equipped with a laser light source, a light receiving element, a scanning optical system, and a signal processing circuit 12, the detection area information storage unit 30, and the obstacle detection unit 40, and the detection area setting device of the present invention is made up of the detection area setting unit 20. The detection area information storage unit 30 stores and manages the detection area. The obstacle detection unit 40 determines the presence or absence of an obstacle in the detection area and outputs information on the presence or absence of an obstacle.

The signal processing circuit 12 is configured with a PLC (programmable logic controller) that controls the laser light source, the light receiving element, and the scanning optical system, and a dedicated DSP (digital signal processor) that performs distance measurement calculations on the reflected light detected by the light receiving element. In this embodiment, an FPGA (field programmable gate array) is used as the signal processing circuit 12.

The detection area information storage unit 30 is composed of a semiconductor memory, and the detection area setting unit 20 and the obstacle detection unit 40 are each composed of a microcomputer and a semiconductor memory in which an application program executed by the microcomputer is stored, an input/output circuit, a communication circuit, etc.

The system control unit 50 is composed of a general-purpose computer installed with application programs for operation management that perform the following processes: control the opening and closing of the door DR of the safety fence SF; detect the stoppage of the vehicle 1; detect the open/closed state of the vehicle door 2; and determine whether the vehicle 1 can depart.

As shown in FIG. 7, when vehicle 1 is not entering platform P, the system control unit 50 sets a general detection area that is independent of the model of vehicle 1, etc. (SD1), detects whether or not an obstacle is present within the general detection area via the obstacle detection unit 40, and if it determines that an obstacle is present (SD2, yes), it outputs an obstacle detection signal (obstacle present) to an external system such as an alarm system (SD3). If no obstacle is detected (SD2, no), it does not output an obstacle detection signal (obstacle present) and proceeds to step SD4, which is detection of vehicle 1 entering the station.

The general-purpose detection area is the smallest detection area set based on the type of vehicle in which the side of the vehicle 1 facing the platform P is closest to the lidar 10. This general-purpose detection area is defined as an initial setting when the system is installed. In addition, it is desirable for the general-purpose detection area to be the smallest detection area among the individual detection areas described in detail below, or a detection area smaller than that.

The system control unit 50 repeats the obstacle detection process (SD2) (SD4, N) until it detects that the vehicle 1 has entered the platform P and stopped based on the input from the entrance detection sensor. When it detects that the vehicle 1 has entered the platform P and stopped (SD4, Y), it starts the detection area setting unit 20 with the door 2 of the vehicle 1 in the closed state before it is opened, and requests the detection area setting unit 20 to set a detection area for the vehicle 1 (SD5). An imaging device or a lidar can be used as the entrance detection sensor, and the function can also be incorporated into the obstacle detection system 100. The detection area is set for each door DR installed in the safety fence SF. This is to detect passengers passing between the door DR and the door 2 of the vehicle 1.

When the system control unit 50 receives a signal from the detection area setting unit 20 indicating that the detection area setting has been completed (SD6, Y), it opens the gate door DR and the door 2 of the vehicle 1, and monitors the movement of passengers getting on and off within the detection area via the detection output of the obstacle detection unit 40 based on the detection area corresponding to the vehicle 1 (SD7).

The system control unit 50 acquires obstacle detection information via the obstacle detection unit 40. If the system control unit 50 determines that an obstacle is present (SD8, present), it outputs an obstacle detection signal (obstacle present) to an external system, etc. (SD11). On the other hand, if the system control unit 50 determines that no obstacle is present (SD8, no), it returns to step SD7 without outputting an obstacle detection signal (obstacle present) and continues the process of repeating the monitoring of passengers getting on and off for the allowable stopping time of the vehicle 1.

When it is time for vehicle 1 to depart, the system control unit 50 determines whether or not it is possible for the vehicle to depart (SD9). The system control unit 50 determines whether or not it is possible for the vehicle to depart based on whether or not there is any risk of a serious accident occurring due to the departure of vehicle 1, such as whether the doors 2 of vehicle 1 are closed and whether or not any passengers or luggage have been detected within the detection area by the obstacle detection process.

If a passenger's body or bag is caught in door 2, or if passengers getting on or off are detected within the detection area, the system control unit 50 reopens door 2 of vehicle 1. Alternatively, the system control unit 50 controls the vehicle to prohibit departure while keeping gate door DR open, and repeats the processing from step SD7 onwards.

If no passengers or luggage are detected within the detection area, the system control unit 50 closes the gate door DR and allows the vehicle 1 to depart (SD9, OK). The system control unit 50 then switches the detection area to a general-purpose detection area (SD10). After this, the system control unit 50 repeats the processes from step SD2 onwards.

In this embodiment, an example has been described in which the system control unit 50 automatically controls the running of the vehicle 1 and the opening and closing of the door 2. However, in a configuration in which the running of the vehicle 1 and the opening and closing of the door 2 are controlled by the driver of the vehicle 1, the system control unit 50 only needs to detect the stop of the vehicle 1 and the open/closed state of the door 2 with a dedicated sensor, and can be configured to output a departure prohibition signal or permission signal to the vehicle 1.

[Configuration of detection area setting device]
3(a) and 3(b), the detection area setting device, i.e., the detection area setting unit 20, includes the functional blocks of a point cloud data acquisition unit 21, a surface acquisition unit 22, a labeling unit 23, and an area setting unit 24. These functional blocks are embodied by executing the above-mentioned application program on a microcomputer.

The detection area setting unit 20 sets a detection area for each Rider 10 installed near the door DR of each gate based on a detection area setting request from the system control unit 50. The detection area setting unit 20 associates each set detection area with an ID previously set for the Rider 10 and stores it in the detection area information storage unit 30. The detection area setting unit 20 then outputs a signal to the system control unit 50 indicating that setting of the detection area has been completed. This allows the obstacle detection unit 40 to identify an individual detection area for each Rider 10. Below, the detection area setting process for one Rider 10 will be described, but it goes without saying that a similar setting process is performed for each Rider 10.

The point cloud data acquisition unit 21 activates the LIDAR 10 and acquires distance image data including distances and directions of distance measurement points including reflection points on the surface of the reference object based on the reflected light of the measurement light irradiated onto an area including the reference object to be set as the detection area. The point cloud data acquisition unit 21 then executes a point cloud data acquisition step of acquiring three-dimensional point cloud data, which are the three-dimensional coordinates of the distance measurement points, by calculation based on the distance image data (FIG. 3(b), S1).

In this embodiment, the detection area is an area defined between a safety fence SF installed on the platform P of the station building and a vehicle 1 parked on the platform P. Furthermore, the "reference object to be measured for which the detection area is to be set" refers to the vehicle 1 and the platform P, and the "area including the reference object to be measured for which the detection area is to be set" refers to an area including the vehicle 1 and the platform P, and including no obstacles such as passengers 3 getting on and off or luggage. Note that in the point cloud data acquisition step, the presence of only a small number of obstacles in the area is acceptable.

In Figure 8(a), an example of the point cloud data obtained at this time is displayed as dots. To make it easier to understand, in Figure 8(a), auxiliary lines in the form of two-dot chain lines are added to the areas corresponding to the side of the vehicle 1 and the floor surface of the platform P.

The scanning range of the measurement light by the LIDAR 10 is set to a range at least larger than the detection area to be set.

Based on the point cloud data acquired by the point cloud data acquisition unit 21, the surface acquisition unit 22 executes a surface acquisition step to acquire multiple surfaces including the surface of the reference object to be measured, specifically the side surface of the vehicle 1 and the floor surface of the platform P (FIG. 3(b), S2).

The labeling unit 23 executes a labeling step in which the multiple surfaces acquired by the surface acquisition unit 22 are labeled under predetermined conditions (FIG. 3(b), S3).

The area setting unit 24 executes an area setting step to set a detection area for detecting obstacles based on the multiple faces labeled by the labeling unit 23 (FIG. 3(b), S4).

Each step is described in detail below.
The LIDAR 10 outputs direction and distance data to the reflection points to the point cloud data acquisition unit 21 via the signal processing circuit 12 each time a unit scan in the horizontal and vertical directions is completed for an area including a reference object to be set as a detection area. When the point cloud data acquisition unit 21 receives this information from the signal processing circuit 12, it stores the point cloud data converted into coordinates in a three-dimensional coordinate space of X, Y, and Z of each reflection point together with a time stamp in an internal storage unit. The three-dimensional coordinate space may be a relative coordinate system with the installation position of each LIDAR 10 as the origin, or may be an absolute coordinate system that is not dependent on the installation position of each LIDAR 10. The point cloud data is not limited to three-dimensional coordinates of X, Y, and Z, and polar coordinates may also be adopted.

Figure 4 shows the process executed by the surface acquisition unit 22. The surface acquisition unit 22 extracts a point of interest and points adjacent to the point of interest from multiple distance measurement points included in the point cloud data, applies principal component analysis (PCA) to these extracted points to generate a surface of interest (SA1), and executes a process (SA2) for all point cloud data to acquire a normal vector to the surface of interest as a normal vector at the point of interest (SA3).

The point of interest refers to point cloud data selected based on a predetermined rule in order to obtain normal vectors from all point cloud data. For example, the thinning rate can be set as the rule. When the thinning rate is set to 50%, each set of point cloud data extracted every other point cloud data in the horizontal and vertical directions from all point cloud data obtained by unit scans in the horizontal and vertical directions becomes the point of interest. The thinning rate is a value that is appropriately set based on the number of all point cloud data, taking into account the calculation load. Alternatively, each point cloud data whose distance is distributed in a predetermined range from among all point cloud data may be set as the point of interest. A rule may be defined by combining both such distance and thinning rate.

The surface acquisition unit 22 searches for distance measurement points within a radius of several mm as neighboring points centered on the point of interest using the kd-tree method. After calculating the center of gravity of the searched point cloud data, the surface acquisition unit 22 performs principal component analysis of the point cloud data. The surface acquisition unit 22 uses the third principal component vector obtained by the principal component analysis as a normal vector (a basis vector with a magnitude of 1) and calculates a plane passing through the center of gravity of the point cloud data previously obtained. Figure 8(b) shows the normal vector obtained for the point cloud data corresponding to Figure 8(a). Note that the point cloud data shown in Figures 8(b) and (c) is data viewed from an angle different from that of Figure 8(a).

In addition to the principal component analysis method mentioned above, other methods for finding a surface from a point of interest and nearby points include the three-dimensional Hough transform method, the least squares method, and the least median method. For example, the least squares method calculates a plane that passes through the center of gravity of each point cloud data and minimizes the squared distance of each point cloud data.

Next, the surface acquisition unit 22 executes a clustering process, which includes the following steps SA4 to SA10.
In the clustering process, first, a specific attention point is extracted from a plurality of attention points, and other attention points existing in a predetermined search range are extracted from the extracted specific attention point (SA4). Then, the specific attention point is compared with the normal vectors of the other attention points (SA5). If the normal vectors of the other attention points fall within a predetermined allowable range of inclination with respect to the direction of the normal vector of the specific attention point (SA6), a clustering process is executed to group the other attention points with the specific attention point (SA7). As candidate points constituting the same surface as the surface containing the specific attention point, the other attention points are grouped in the same loop as the specific attention point. In this embodiment, the inclination of the predetermined allowable range is set to 1°. The inclination of the predetermined allowable range is a value that can be appropriately set according to the object and the like. In addition, when grouping, the attention point and the distance measurement point that is a neighboring point of the attention point used to calculate the normal vector of the attention point may be grouped together, or only the attention point may be grouped together. In the latter case, the normal vector may be included as an attribute value of the attention point.

The surface acquisition unit 22 repeats the clustering process with each of the points of interest as a specific point of interest (SA4 to SA10). If any of the points of interest to be grouped in the subsequent clustering process have already been grouped in the previous clustering process (SA8, Y), the surface acquisition unit 22 executes an integration process to integrate all of the points of interest to be grouped in the subsequent clustering process into the same group as the previous clustering process (SA9).

In this embodiment, a space with a radius of about 2 m centered on a specific point of interest is set as the specified search range. For example, there are few measurement points above and below the window frame of the door 2 of the vehicle 1, and the estimation accuracy of normals is low, so there is a risk that the left and right sides of the vehicle 1, separated by the window, will be clustered as separate surfaces. Even in such cases, by setting a space with a radius of about 2 m as the search range, the left and right sides of the vehicle 1, separated by the window, will be clustered as the same surface. The value of the specified search range can also be set appropriately depending on the target object, etc.

Furthermore, after executing the integration process in step SA9, the surface acquisition unit 22 executes a second integration process (SA10'). The surface acquisition process obtains a representative normal vector representing each of the multiple groups generated by the clustering processes in steps SA4 to SA10, and integrates groups whose respective representative normal vectors fall within a predetermined tolerance range into the same group.

The points of interest grouped by the clustering process in steps SA4 to SA10 are a collection of multiple points of interest that exist within a specified search range based on at least a specific point of interest. If the search range is set to a large value, points of interest that would normally belong to different groups will end up in the same group, so the search range must be limited to a certain size, for example a space with a radius of about 2 m as mentioned above. If the search range is set to a smaller value than this, points of interest that should normally belong to the same group may end up in different groups. Even in such cases, they can be merged into the same group, provided that the representative normal vectors of each group fall within a specified tolerance range.

The vector that is the median or average of the orientations of the normal vectors of the points of interest that belong to the same group can be used as the representative normal vector. An approximate plane can be generated from multiple points of interest that belong to the same group, and a basis vector perpendicular to that plane can be used as the representative normal vector.

The specified tolerance range can be defined, for example, by the conditions that the orientation of the representative normal vector falls within a specified tolerance range, and that the distance between an approximate plane generated from multiple attention points belonging to the same group and an approximate plane generated from multiple attention points belonging to another group falls within a specified range.

The second integration process of step SA10' is a selective process, and if the search range is set to an appropriate value, the second integration process is not essential. In addition, the integration processes of steps SA8 and SA9 described above are also selective, and if the integration processes of steps SA8 and SA9 are not executed, the second integration process is executed to effectively integrate groups that belong to the same surface.

If the search range is enlarged in steps SA8 and SA9, or the allowable range of the distance between faces is increased in step SA10', and the normal vector has a slope within a predetermined allowable range, a point of interest that is far from a face that should be composed of points of interest included in the same group may be combined into the same group. If this occurs, for example, there is a risk that a distance measurement point based on light reflected by the inside wall of the vehicle after passing through the window glass may be included in the same group.

The surface acquisition unit 22 then performs a filtering process on the point cloud data that has been grouped into the same group by the clustering process, thereby removing outliers from the same group (SA11). This process appropriately removes distance measurement points based on light that has passed through the window glass and is reflected by the inner wall of the vehicle. Here, an outlier is a value that has an abnormally large residual from an estimate of the true value among the obtained observation values.

As a filtering process, RANSAC is preferably used. In this embodiment, three points are randomly selected from the points of interest contained in the same group, and the distance between the plane created with those three points and each of the other points is calculated. If there are a certain number (1/4 of the total) or more of points whose calculated distance is equal to or less than a predetermined threshold (0.1 m), the points of interest are left as candidates, excluding outliers that deviate from the threshold. The surface acquisition unit 22 executes this process a predetermined number of times, and uses the point with the largest number of points that are equal to or less than the threshold to calculate the equation of the plane of the cluster. Note that the number of points selected to create the above-mentioned plane, the threshold, etc. are not particularly limited, and can be set as appropriate.

As a filtering process for removing outliers, in addition to RANSAC, M-estimation or the least median method can be used, and it is preferable to use a process that performs robust estimation including these. In FIG. 8(c), the outliers removed from the cluster representing the floor surface of platform P are shown in the dashed circle.

Next, after applying a filtering process to each group, a predetermined number of faces are extracted in descending order of the number of distance measurement points (SA12). In this embodiment, the side of the vehicle 1 and the floor surface of the platform P are obtained by extracting two faces in descending order of the number.

Figure 5 shows the process executed by the labeling unit 23. The labeling unit 23 distinguishably labels a reference plane corresponding to the surface of the reference object from among the multiple surfaces and an intersecting plane that intersects the reference plane, based on a preset reference vector and the normal vectors of the multiple surfaces acquired by the surface acquisition unit 22.

In this embodiment, a vertical base vector is set as the reference vector. Of the two surfaces extracted in step SA2, one surface, i.e., a surface whose normal vector is orthogonal to the reference vector and that runs along the side of the vehicle 1 (the side along the direction of travel of the vehicle 1 and that is close to the platform P), is identified as the first surface (SB1). The other surface, i.e., a surface whose normal vector is parallel to the reference vector, is identified as the second surface that runs along the floor surface of the platform P (SB2). Figure 9 shows the reference vector VB, the normal vector VT of the first surface, the normal vector VP of the second surface, the first surface C1, and the second surface C2.

The labeling unit 23 determines that the normal vector is orthogonal when the inclination angle of the normal vector with respect to the reference vector is within the range of 90°±Δθ1, and determines that the normal vector is parallel when the inclination angle of the normal vector with respect to the reference vector is within the range of 0°±Δθ2. The values of Δθ1 and θ2 are not particularly limited, but are set to Δθ1=20° and Δθ2=10° in this embodiment.

Figure 6 shows the process executed by the area setting unit 24. The area setting unit 24 defines a detection area consisting of a polyhedron on the side of the measurement light irradiation source (lidar 10 side) based on the reference surface (first surface) labeled by the labeling unit 23 and the intersecting surface (second surface) that intersects with the reference surface (first surface). The explanation will continue with reference to Figures 9 to 11.

First, the area setting unit 24 identifies the directions of the normal vectors VT and VP that have already been calculated for the first surface C1 and the second surface C2 (SC1). The area setting unit 24 calculates the vector product (cross product) VT×VP of both vectors VT and VP to calculate the intersection vector between the first surface C1 and the second surface C2 and obtain the intersection line Li (SC2).

Next, the area setting unit 24 sets a width W that is the widthwise boundary of the detection area in the left-right direction along the intersection line Li, based on the center point CP on the intersection line Li that is the shortest distance from the rider 10 (SC3).

Next, the area setting unit 24 sets a depth D that is the depth-direction boundary of the detection area from the intersection line along the second surface (SC4), and sets a height H that is the height-direction boundary from the second surface (SC5). As a result, as shown in FIG. 10, a rectangular surface S1 with a width W and height H is formed on the first surface C1, and a rectangular surface S2 with a width W and depth D is formed on the second surface C2.

Next, the area setting unit 24 generates four auxiliary surfaces: auxiliary surfaces S3 and S4 that face the rectangular surfaces S1 and S2, and auxiliary surfaces S5 and S6 that are perpendicular to the rectangular surfaces S1 and S2 (SC6).

As shown in FIG. 11, the area setting unit 24 shifts the rectangular surface S1 corresponding to the side of the vehicle 1 in the depth direction by an offset amount ΔD for the reference detection area BA obtained in this manner. In addition, the rectangular surface S2 corresponding to the floor surface of the platform P is shifted upward by an offset amount ΔH. The area setting unit 24 defines the rectangular parallelepiped region obtained in this manner as the final detection area DA (SC7). The area setting unit 24 stores the coordinates of each vertex indicating the detection area DA in the detection area information storage unit 30 (SC8).

In this way, by setting a detection area DA for each vehicle 1 that enters the platform P, it is possible to set a detection area DA with as few blind spots as possible to accommodate various vehicles 1 with different widths. Note that the offset values ΔD and ΔH described above are set appropriately so as not to erroneously detect the side walls of the vehicle or the floor of the platform P as an obstacle. If the offset values ΔD and ΔH are set large, the blind spots become large, and if the offset values ΔD and ΔH are set small, the probability of erroneously detecting an obstacle increases.

In the above-described embodiment, the detection area setting unit 20 defines the reference detection area BA each time the vehicle 1 enters the platform P, and further defines the final detection area DA based on the offset values ΔD and ΔH. However, the reference detection area BA may be defined only when the vehicle 1 first enters the platform, and the detection area may be set by using the data of the reference detection area BA when the vehicle 1 subsequently enters the platform. Specifically, based on the data of the reference detection area BA initially acquired for each of the rectangular surface S2 corresponding to the floor surface of the platform P and the auxiliary surfaces S3 to S6, the surface acquisition unit 22 acquires only the first surface corresponding to the side surface of the vehicle 1, and generates a new rectangular surface S1.

In the above embodiment, an example has been described in which the surface acquisition unit 22 extracts two surfaces in descending order of the number of distance measurement points (points of interest), that is, the surfaces corresponding to the side of the vehicle 1 and the floor surface of the platform P, but if measurable, the surface corresponding to the safety fence SF may also be extracted. In this case, the area setting unit 24 does not need to generate the auxiliary surface S3, and an offset toward the vehicle 1 is set on the surface corresponding to the safety fence SF. In addition, the shape of the detection area DA is not limited to a rectangular parallelepiped, and it may also be configured as a hexahedron or even a polyhedron.

After the door 2 of the vehicle 1 is opened, the obstacle detection unit 40 determines whether the data for the distance measurement point output from the signal processing circuit 12 of the LIDAR 10 is data within the detection area DA, and if the data is within the detection area DA, it determines that an obstacle is present.

2, the detection area setting unit 20, the detection area information storage unit 30, the obstacle detection unit 40, and the signal processing circuit 12 are each configured separately from the rider 10 and connected by communication lines. However, the signal processing circuit 12, the detection area setting unit 20, the detection area information storage unit 30, and the obstacle detection unit 40 may each be integrally assembled within the housing of the rider 10, or only the signal processing circuit 12 may be assembled within the housing of the rider 10. Alternatively, the signal processing circuit 12, the detection area information storage unit 30, and the obstacle detection unit 40 may be assembled within the housing of the rider 10, and the detection area setting unit 20 may be configured separately.

In the above-described embodiment, an example was described in which the system control unit 50 sets a general-purpose detection area that is not dependent on the model of the vehicle 1 as the minimum detection area when the vehicle 1 is not at the platform P. However, the general-purpose detection area needs to be individually defined as an appropriate range as an initial setting when the system is installed, and separate work such as preparatory work is required for this.

Therefore, when vehicle 1 is not entering platform P, the system control unit 50 requests the detection area setting unit 20 to set a general-purpose detection area, and the detection area setting unit 20 is configured to automatically generate the general-purpose detection area, thereby making it possible to eliminate the need for cumbersome preparatory work. Below, the process for setting the general-purpose detection area by the detection area setting unit 20 will be described in detail, focusing on the differences from the detection area setting process described above.

In this example, the "reference object to be measured for which a detection area is to be set" refers to the platform P, and the "area including the reference object to be measured for which a detection area is to be set" refers to an area including the platform P and free of obstacles such as boarding and alighting passengers 3 or baggage, including the track surface of the vehicle 1.

When the system control unit 50 receives a command to set a general-purpose detection area from the system control unit 50 while the vehicle 1 is not entering the platform P, the point cloud data acquisition unit 21, surface acquisition unit 22, labeling unit 23, and area setting unit 24 provided in the detection area setting unit 20 execute the setting process of the general-purpose detection area.

As shown in FIG. 12, the point cloud data acquisition unit 21 acquires three-dimensional point cloud data indicating the coordinates of distance measurement points including reflection points on the surface of the reference object to be measured, based on the reflected light of the measurement light irradiated onto the reference object to be set as the general-purpose detection area, i.e., an area including the platform P (SE1).

When the point cloud data is acquired by the point cloud data acquisition unit 21, the surface acquisition unit 22 extracts a point of interest and points adjacent to the point of interest from the multiple distance measurement points included in the point cloud data, applies principal component analysis to these extracted points to generate a surface of interest (SE2), and executes a process (SE3) for all the point cloud data to acquire a normal vector to the surface of interest as a normal vector at the point of interest (SE4).

Next, the surface acquisition unit 22 executes a clustering process (SE5 to SE11). This clustering process is similar to the clustering process (SA4 to SA10) previously described with reference to FIG. 4, and therefore a detailed description will be omitted. However, since a situation does not arise in which the left and right sides of the vehicle 1, separated by a window, are clustered into the same surface, as in the previous example, the search range can be smaller than in the previous example, and can be set to a space with a radius of about 0.1 m centered on a specific point of interest, for example. Furthermore, the surface acquisition unit 22 removes outliers from the same group by executing a filtering process on the point cloud data grouped into the same group in the clustering process (SE12).

Figure 13(a) shows point cloud data mainly corresponding to the platform P and the track surface detected by the lidar 10. Figure 13(b) shows normal vectors obtained for the point cloud data corresponding to Figure 13(a).

Returning to FIG. 12, the explanation will continue. The labeling unit 23 identifies a reference plane RP corresponding to the surface of the platform P, which is the reference object to be measured, from among the multiple planes based on a preset reference vector and the normal vectors of the multiple planes acquired by the plane acquisition unit 22 (SE13).

As a reference vector, a vertical base vector detected by an acceleration sensor built into the rider 10 is set as the reference vector VB, and among multiple planes having normal vectors whose direction coincides with that of the reference vector VB, the plane having the closest vertical distance from the rider 10 is identified as the reference plane RP. If the inclination angle of the normal vector with respect to the reference vector VB is within the range of 0°±Δθ2, it is determined to be parallel. The value of Δθ2 is not particularly limited, but in this embodiment it is set to Δθ2=10°. Note that the setting of the reference vector is not limited to a mode using an acceleration sensor built into the rider 10, and may be a mode in which data is input in advance corresponding to the installation posture of the rider 10, for example.

Next, the labeling unit 23 acquires the boundary line BL of the reference plane RP based on the point cloud data acquired by the point cloud data acquisition unit 21 (SE14). As shown in FIG. 14, the labeling unit 23 captures distances d0, d1, ..., dn, dn+1, ..., dm (n and m are integers) obtained along the scanning direction (indicated by an arrow in FIG. 14) of the first deflection mechanism 14A of the laser light emitted from the lidar 10, and repeats the process of obtaining a boundary point where the value changes suddenly (in FIG. 14, the position corresponding to the distance dn (dn<<dn+1)) along the scanning direction of the second deflection mechanism 14B, and acquires an approximation line passing through the obtained multiple boundary points as the boundary line BL. As shown in FIG. 14, the boundary line BL acquired in this embodiment corresponds to the end (edge) of the platform P, which is the reference object to be measured, on the track surface side. Since there is a difference in height between the platform P and the track surface, it is possible to obtain a boundary where the value of the distance dn changes suddenly.

Furthermore, as shown in FIG. 15(a), the labeling unit 23 generates a cross plane CRP (in this embodiment, the cross angle is set to 90°, making it a vertical plane) that passes through the boundary line BL and crosses the floor surface of the platform P, which is the reference plane RP, and identifies the reference plane RP and the vertical plane that is the cross plane CRP perpendicular to the reference plane RP, and labels each of them (SE15). Note that the size of the cross plane CRP is not particularly limited and is arbitrary. Also, the cross angle is not limited to 90°, and may be set to any range within which a foreign object present on the reference plane RP can be detected, for example, within a range of about 90±5°.

Next, as shown in FIG. 15(b), the area setting unit 24 sets a width W, which is the widthwise boundary of the general-purpose detection area, in the left-right direction along the boundary line BL, based on the center point CP on the boundary line BL that is the shortest distance from the rider 10 (SE16).

Next, the area setting unit 24 sets a depth D that is the depth-direction boundary of the general-purpose detection area from the boundary line BL along the reference plane RP (SE17), and sets a height H that is the height-direction boundary from the reference plane RP (SE18). As a result, a rectangular surface S1' with width W and height H is formed on the plane of the cross plane CRP, and a rectangular surface S2' with width W and depth D is formed on the plane of the reference plane RP.

Next, the area setting unit 24 generates four auxiliary surfaces: auxiliary surfaces S3', S4' that face the rectangular surfaces S1', S2', and auxiliary surfaces S5', S6' that are perpendicular to the rectangular surfaces S1', S2' (SE19), and defines a general-purpose detection area BA' that is surrounded by the six surfaces S1', S2', S3', S4', S5', S6' (SE20).

As shown in FIG. 15(c), the area setting unit 24 shifts the rectangular surface S1' corresponding to the crossing plane CRP in the depth direction by an offset amount ΔD for the general-purpose detection area BA' obtained in this way. In addition, the rectangular surface S2' corresponding to the reference plane RP is shifted upward by an offset amount ΔH. The area setting unit 24 defines the rectangular parallelepiped region obtained in this way as the final general-purpose detection area DA' (SE21). The area setting unit 24 stores the coordinates of each vertex indicating the general-purpose detection area DA' in the detection area information storage unit 30. Note that the values of the offset amounts ΔD and ΔH are not particularly limited, and may be set to values different from those described in FIG. 11. Therefore, in this embodiment, the final general-purpose detection area DA' can be defined by acquiring the boundary line BL corresponding to the end of the platform P on the track surface side.

In the above embodiment, an example was described in which the surface acquired by the surface acquisition unit 22 was a flat surface, but the present invention is not limited to flat surfaces, and may also be applied to curved surfaces with small curvature. For example, a known surface fitting algorithm may be used, in which case the normal of the tangent plane to the point of interest may be adopted as the normal of the surface of interest.

In the above-described embodiment, an example was described in which the detection area is an area defined between a safety fence SF installed on a platform P of a station building and a vehicle 1 parked on the platform P. However, the present invention is not limited to this aspect, and can be used in any facility that requires setting an appropriate detection area in an environment in which the detection area fluctuates.

For example, this can be applied to cases where a car that has entered and stopped in a multi-story parking lot is used as the reference measurement object, and a detection area is set based on the coordinates of distance measurement points including reflection points on the surface of the car, based on the reflected light of measurement light irradiated onto an area including the car before people get in or out of the car. When the transport mechanism that transports the car to the garage is operating, the presence of obstacles such as people or luggage around the car can be confirmed to ensure safety.

The point cloud data acquisition unit 21 is not limited to a method of irradiating light as in the above embodiment. For example, any means capable of acquiring point cloud data, such as a passive stereo camera, ultrasonic wave, or millimeter wave sensor, may be used.

The above-mentioned embodiments are merely examples of the present invention, and the scope of the present invention is not limited to these descriptions. Needless to say, the present invention can be modified and designed as appropriate within the scope of the effects achieved by the present invention.

1: Vehicle 2: Door 10: Lidar 12: Signal processing circuit 20: Detection area setting unit 21: Point cloud data acquisition unit 22: Surface acquisition unit 23: Labeling unit 24: Area setting unit 30: Detection area information storage unit 40: Obstacle detection unit 50: System control unit BA: Reference detection area DA: Detection area DR: Door SF: Safety fence

Claims (20)

a point cloud data acquisition unit that acquires three-dimensional point cloud data indicating coordinates of distance measurement points including reflection points on a surface of a reference object based on reflected light of measurement light irradiated onto an area including the reference object to be set as a detection area;
a surface acquisition unit that acquires a plurality of surfaces including a surface of the reference object based on the point cloud data acquired by the point cloud data acquisition unit;
a labeling unit that labels the multiple faces acquired by the face acquisition unit under a predetermined condition;
an area setting unit that sets the detection area for detecting an obstacle based on the plurality of faces labeled by the labeling unit;
The present invention is configured to include
The surface acquisition unit executes a process for acquiring, for a plurality of focus points, normals to a surface of interest generated at a focus point and a point adjacent to the focus point among a plurality of distance points included in the point cloud data, as normal vectors at the focus points, and executes a surface acquisition process for acquiring a plurality of surfaces including any of the focus points based on each normal vector. The detection area setting device according to claim 1, wherein the surface acquisition process includes a clustering process in which, if the normal vectors of other points of interest that exist within a predetermined search range from a specific point of interest are inclined within a predetermined allowable range with respect to the direction of the normal vector of the specific point of interest, the other points of interest are grouped together with the specific point of interest. The detection area setting device according to claim 2, wherein the surface acquisition process includes a consolidation process in which the clustering process is repeated with each of the points of interest as the specific points of interest, and in the case where any of the points of interest to be grouped in a later clustering process has already been grouped in a previous clustering process, all of the points of interest to be grouped in the later clustering process are consolidated into the same group as the points of interest in the previous clustering process. The detection area setting device according to claim 2 or 3, wherein the surface acquisition process includes a consolidation process in which a representative normal vector representing each of the multiple groups generated by the clustering process is obtained, and groups whose respective representative normal vectors fall within a predetermined tolerance range are consolidated into the same group. The detection area setting device according to any one of claims 2 to 4, wherein the surface acquisition process performs a filtering process on the points of interest grouped into the same group to remove outliers from the same group. The detection area setting device according to claim 5, wherein the filtering process is a process that performs robust estimation including any of RANSAC, M-estimation, and least median methods. The detection area setting device according to any one of claims 2 to 6, wherein the surface acquisition process acquires a surface based on points of interest included in the same group. The detection area setting device according to any one of claims 1 to 7, wherein the labeling unit identifies and labels a reference plane corresponding to the surface of the reference object from among the multiple surfaces and an intersecting plane that intersects with the reference plane based on a preset reference vector and normal vectors of the multiple surfaces acquired by the surface acquisition unit. The detection area setting device according to any one of claims 1 to 7, wherein the labeling unit identifies a reference surface corresponding to the surface of the reference object from the multiple surfaces based on a preset reference vector and the normal vectors of the multiple surfaces acquired by the surface acquisition unit, acquires a boundary line of the reference surface based on the point cloud data acquired by the point cloud data acquisition unit, generates an intersecting surface that passes through the boundary line and intersects with the reference surface, and identifies and labels the reference surface and the intersecting surface that intersects with the reference surface. The detection area setting device according to claim 8 or 9, wherein the area setting unit defines a detection area consisting of a polyhedron between the reference surface and the irradiation source of the measurement light based on the reference surface and the intersection surface labeled by the labeling unit. A detection area setting device according to any one of claims 1 to 10,
a range sensor that scans a measurement light in an area including the detection area set by the detection area setting device and determines the presence or absence of an obstacle based on the presence or absence of reflected light from the detection area;
An obstacle detection system comprising: a point cloud data acquisition step of acquiring three-dimensional point cloud data indicating coordinates of distance measurement points including reflection points on a surface of a reference object based on reflected light of measurement light irradiated onto an area including the reference object to be set as a detection area;
a surface acquiring step of acquiring a plurality of surfaces including a surface of the reference object based on the point cloud data acquired in the point cloud data acquiring step;
a labeling step of labeling the plurality of faces acquired in the face acquisition step under a predetermined condition;
an area setting step of setting the detection area for detecting an obstacle based on the plurality of faces labeled in the labeling step;
The present invention is configured to include
The surface acquisition step is a detection area setting method characterized in that a process of acquiring, for a plurality of attention points, normals to a surface of interest generated at a point of interest and a point near the point of interest among a plurality of distance points included in the point cloud data as normal vectors at the point of interest, and acquiring a plurality of surfaces constituted by any of the attention points based on each normal vector. The detection area setting method according to claim 12, wherein the surface acquisition step includes performing a clustering process in which, if the normal vectors of other points of interest that exist within a predetermined search range from the specific point of interest are inclined within a predetermined allowable range with respect to the direction of the normal vector of the specific point of interest, the other points of interest are grouped together with the specific point of interest. The detection area setting method according to claim 13, wherein the surface acquisition step includes repeating the clustering process with each attention point as the specific attention point, and, if any of the attention points to be grouped in the subsequent clustering process have already been grouped in the previous clustering process, executing an integration process to integrate all attention points to be grouped in the subsequent clustering process into the same group as the previous clustering process. The detection area setting method according to claim 13 or 14, wherein the surface acquisition step includes performing a consolidation process in which a representative normal vector representing each of the multiple groups generated by the clustering process is obtained, and groups whose respective representative normal vectors have orientations within the predetermined tolerance range are consolidated into the same group. The detection area setting method according to any one of claims 13 to 15, wherein the surface acquisition step performs a filtering process on the points of interest grouped into the same group to remove outliers from the same group. The detection area setting method according to claim 16, wherein the filtering process is a process for performing robust estimation including any one of RANSAC, M-estimation, or least median method. The detection area setting method according to any one of claims 12 to 17, wherein the labeling step identifies and labels a reference plane corresponding to the surface of the reference object from among the multiple surfaces and an intersecting plane that intersects with the reference plane based on a preset reference vector and normal vectors of the multiple surfaces acquired in the surface acquisition step. The detection area setting method according to any one of claims 12 to 17, wherein the labeling step identifies a reference surface corresponding to the surface of the reference object from the multiple surfaces based on a preset reference vector and normal vectors of the multiple surfaces acquired in the surface acquisition step, acquires a boundary line of the reference surface based on the point cloud data acquired in the point cloud data acquisition step, generates an intersecting surface that passes through the boundary line and intersects with the reference surface, and identifies and labels the reference surface and the intersecting surface that intersects with the reference surface. 20. The detection area setting method according to claim 18 or 19, wherein the area setting step defines a detection area consisting of a polyhedron between the reference surface and the source of the measurement light based on the reference surface and the intersection surface labeled in the labeling step.

Images