JP2024075139A - Object detection device and object detection method - Google Patents

Abstract

To provide an object detection device and an object detection method capable of coping with a complicated structure and detecting an object by automatically detecting a background area.SOLUTION: An object detection device 10 includes: a cube processing unit 11a installed on the ground to scan a scanning range by intermittently irradiating light, receive point group information from a LiDAR 2 for receiving reflection light of the light to acquire the point group information based on the reflection light, and detect a background area being an area of a stationary object within a prescribed space S on the basis of point group information inputted multiple times during a prescribed time; and a detection unit 11b for detecting an object in the prescribed space S on the basis of point group information included in an area other than the background area detected by the cube processing unit 11a. Then, the cube processing unit 11a detects the background area on the basis of a detection state of a point group in each of a plurality of cubic areas obtained by dividing the prescribed space S by an octree.SELECTED DRAWING: Figure 2

Description

The present invention relates to an object detection device and an object detection method for detecting objects, for example, on a road.

For example, a method is known for an object detection device that detects objects on the road, which detects objects using clusters generated by clustering point clouds detected by LiDAR (Light Detection And Ranging).

For example, Patent Document 1 describes a method for detecting objects in the correct units with high accuracy by dividing an over-connected cluster if it is determined that the cluster is over-connected based on the image target detected from the ambient light image and the cluster obtained by clustering the point cloud, and by combining two or more clusters that exist in a part of the point cloud that corresponds to the image target if it is determined that the cluster is over-divided.

JP 2021-131385 A

As with the method described in Patent Document 1, it is possible to remove the effects of stationary objects (background regions) such as houses and walls by detecting over-coupling, but this has the problem of complicating the processing.

One way to deal with this is to detect only moving objects by detecting the background area in advance and excluding the points contained in that background area. However, if an area with trees is defined as a two-dimensional background area, for example, pedestrians passing under the tree branches may not be detected.

Furthermore, stationary objects that make up the background region are not limited to relatively simple shapes such as walls. For example, vegetation not only has a complex shape to begin with, but also changes in shape due to growth and the presence or absence of leaves depending on the season. There are also environmental changes such as road construction and snowdrifts. Conventionally, such environmental changes have been checked manually and the background region remeasured, but as this is very time-consuming, automation is desirable.

The present invention aims to provide an object detection device and method that can handle complex structures and automatically detect background areas and objects.

The invention described in claim 1, which has been made to solve the above problems, is an object detection device that includes an input unit that receives point cloud information from a sensor that is installed on the ground, intermittently irradiates light to scan a specified space, receives reflected light of the light, and acquires point cloud information based on the reflected light; a detection unit that detects a background region that is an area of stationary objects in the specified space based on the point cloud information input to the input unit multiple times within a specified time; and an object detection unit that detects objects in the specified space based on the point cloud information included in an area other than the background area detected by the detection unit, and the detection unit detects the background region based on the detection state of the point cloud in each of a number of cubic regions obtained by dividing the specified space by an octree.

According to the present invention, a background region is detected based on the detection state of point clouds in each of multiple cubic spaces obtained by dividing a specified space using an octree, so even if the shape of the background region is complex, it can be approximated by a combination of cubes. In addition, by inputting point cloud information, it becomes possible to automatically detect the background region.

1 is an example of installation of an object detection device according to an embodiment of the present invention. FIG. 2 is a block diagram of the object detection device shown in FIG. 1 . 3 is a flowchart of the operation of the object detection device shown in FIG. 2 . FIG. 1 is an explanatory diagram of a polygonal closed region. FIG. 1 is an explanatory diagram of an example of division of a predetermined space. FIG. 2 is an explanatory diagram of an octree structure. 1 is a diagram showing the relationship between the installation positions of sensors and observation points on the xy plane. 11 is an explanatory diagram of a case where the vertex with the shortest distance is not the shortest distance from the reference point of the sensor to the cubic area. FIG. FIG. 11 is an explanatory diagram of the change in size of a cubic region depending on the distance from a sensor. FIG. 11 is an explanatory diagram of a background determination threshold value. FIG. 13 is an explanatory diagram of integrating child nodes.

One embodiment of the present invention will be described below with reference to Figs. 1 to 11. Fig. 1 shows an example of an installation of an object detection device according to one embodiment of the present invention.

The object detection device 10 shown in FIG. 1 is installed on a fixed support pole P, for example, near an intersection. The object detection device 10 detects moving objects such as pedestrians and vehicles passing through the intersection. In other words, the object detection device 10 is installed on the ground. Installed on the ground here means that it is fixed to the ground, either directly or indirectly, that is, it is installed on an object that is fixedly installed on the ground, such as a building, rather than on a moving object such as a vehicle.

The object detection device 10 includes a main body 1 and a LiDAR 2. The main body 1 is provided as a control box or the like at the bottom of the support P. As shown in the block diagram of FIG. 2, the main body 1 includes a control unit 11 and a communication unit 12.

The control unit 11 is composed of, for example, a computer. The control unit 11 is responsible for the overall control of the object detection device 10. In FIG. 1, the control unit 11 functionally comprises a cube processing unit 11a and a detection unit 11b.

As described below, the cube processing unit 11a divides a predetermined space into multiple cube regions based on the point cloud information acquired by the LiDAR2, and determines whether each cube region is a background region or a detection target region. A background region is a stationary object other than a moving object such as a vehicle or a pedestrian, and is an area that is not subject to object detection by the detection unit 11b. The detection unit 11b detects objects based on the background region defined by the cube processing unit 11a.

The communication unit 12 outputs information about the object detected by the control unit 11 to the outside. The communication unit 12 may be configured as, for example, a wireless communication device (it may also be a wired communication device). Information about the object may include the type of object, the direction of movement, and the speed of movement. Specific examples of output destinations include moving objects such as vehicles passing near the support pillar P, and traffic control centers.

LiDAR2 is a sensor that intermittently irradiates laser light to scan a specified space, receives the reflected light of the light, and acquires point cloud information (three-dimensional point cloud information) based on the reflected light. LiDAR2 discretely measures the distance to an object in the space within the irradiation range using the reflected light of the irradiated laser light, and recognizes the position and shape of the object as a three-dimensional point cloud. In this embodiment, LiDAR2 is installed on the top of a pillar P as shown in FIG. 1, and scans the road surface R (road surface R and the space above it) as the specified space.

Next, the operation of the object detection device configured as described above will be described with reference to the flowchart in FIG. 3. First, the control unit 11 acquires observation points from the LiDAR 2 (step S1). That is, the control unit 11 acquires the point cloud information acquired by the LiDAR 2. That is, the control unit 11 functions as an input unit to which the point cloud information is input from the LiDAR 2.

Next, the cube processing unit 11a determines whether the observation point is within a polygonal closed region (step S2). A polygonal closed region will be described with reference to FIG. 4. The left side of FIG. 4 is a diagram showing a predetermined space S including a polygonal closed region CL. In this embodiment, as described later, a cube-shaped predetermined space S including a space that can be scanned by the LiDAR2 is defined, and the predetermined space S is divided into multiple cubic regions using an octree, and this process is repeated recursively. However, in a realistic usage environment of the LiDAR2, particularly when a road is assumed and distant vehicles, etc. are to be detected, there will be a large amount of excess space in the width and height directions compared to the depth direction. This excess space is defined in advance as a background region and is excluded from the process of detecting the background region described later.

Specifically, a planar polygon is defined by multiple vertex coordinates (x, y) on a width (x)-depth (y) plane (top right in Fig. 4). Next, the planar polygon is expanded in the height (z) direction within the range of z _min to z _max to define a three-dimensional closed region (bottom right in Fig. 4). If the observation point is inside the closed region, it is subject to cubic region division, which will be described later, and to background region inside/outside determination using the same region, and if it is outside the closed region, it is excluded from subsequent processing as a background region. The polygonal closed region CL shown in Fig. 4 is set in advance in the control unit 11 before executing the flowchart in Fig. 3.

In addition, although a quadrangle is used as the polygon in Fig. 3, it is not limited to a quadrangle. However, a quadrangle is preferable for convenience of dividing into cubic regions. The planar polygon and _zmin to _zmax described above may be determined based on the results of manual measurements at the installation position of the object detection device 10. That is, the cube processing unit 11a sets an area within a predetermined space S, the range of which is limited in the horizontal plane and height directions, and detects a background area within the limited area.

Next, the cube processing unit 11a determines whether a cube region has not yet been generated (step S3). If a cube region has not yet been generated (step S3; YES), the size of the cube region is determined (step S4). On the other hand, if a cube region has not yet been generated (step S3; NO), the process proceeds to step S6, which will be described later.

The space division in this embodiment will now be described with reference to Figures 5 and 6. Note that in Figure 5, for simplicity's sake, a cubic space S is used as a given space for which step S2 has not been performed, but in reality, the following processing is performed within the polygonal closed region described above.

Figure 5 is an explanatory diagram of an example of the division of a specified space S in this embodiment. The specified space S shown in Figure 5 includes a space that can be scanned by the LiDAR 2. In this embodiment, the specified space S is defined as a cube, and the cube is divided by an octree. By dividing recursively by an octree, the specified space S can be divided into minute cubic regions, for example, 10 cm square.

Figure 6 shows an octree structure. As is well known, an octree is a tree structure in which one parent node has eight child nodes. The octree structure can quickly reference any node (cubic region) using a well-known method called Morton order. This octree structure is formed in the memory of the control unit 11, and each node stores information indicating whether the node (cubic region) is a background region or not, as well as the values of various counters, which will be described later.

In step S3, it is determined whether a cubic region has already been generated by the processes executed in steps S4 and S5.

Next, the processing of step S4 will be described with reference to Figs. 7 and 8. Since the spatial resolution of LiDAR2 decreases as the distance increases, applying a cubic region size suitable for the vicinity of LiDAR2 to a location far from LiDAR2, as shown in Fig. 7, results in excessive region division.

Fig. 7 is a diagram showing the relationship between the installation position of the LiDAR 2 and the observation points on the xy plane. In Fig. 7, the symbol L indicates the installation position of the LiDAR, the symbols P ₁₁ and P ₁₂ indicate the observation points at a distance l ₁ from the installation position L of the LiDAR 2, and the symbols P ₂₁ and P ₂₂ indicate the observation points at a distance l ₂ (l ₁ <l ₂ ) from the installation position L of the LiDAR 2. The observation points P ₁₁ and P ₁₂ and the observation points P ₂₁ and P ₂₂ are adjacent observation points (continuous pulsed light). Here, if the cubic region is divided into the sizes indicated by symbols Ca1 and Cb1, observation points _P11 and _P12 may be observed in the cubic regions Ca1 and Cb1 at distance _l1 , but observation points _P21 and _P22 will not be observed in the cubic regions Ca2 and Cb2 of the same size at distance _l2 . In other words, the cubic region is divided excessively at distance _l2 .

Therefore, in this embodiment, the number of recursive divisions using an octree is changed depending on the distance from LiDAR2 to the cubic region. Specifically, for the target cubic region, the distance from the reference point (installation position) of LiDAR2 to the vertex with the shortest distance is calculated, and the distance between observation points in the width direction at that position is calculated. The distance between observation points can be calculated, for example, from the irradiation interval of the laser light and the operating speed (frequency) of a scanning means such as a mirror.

In the case of Figure 7, the reference point is the installation position L of LiDAR2, the vertex with the shortest distance is the vertex Pv that is the boundary between cubic regions Ca1 and Cb1 and is closest to the installation position L, and the distance between the observation points in the width direction corresponds to the interval lx between the observation points in the x direction at vertex Pv.

In the case of Fig. 7, the angle between the line connecting the installation position L and the vertex Pv and the x-axis is nearly a right angle (almost the shortest distance), so the distance between the installation position L and the vertex Pv can be used as is, but there are cases where the vertex with the shortest distance is not the shortest distance from the reference point of the sensor to the cubic region, as shown in Fig. 8. In this case, the vertex with the shortest distance and the intersection point _P3 with the shortest distance from the reference point of the sensor to the cubic region can be found and used to calculate the distance between the observation points.

Next, half the distance between the observation points in the width direction is set as the threshold value for the size of the cubic region. In this embodiment, this threshold value is set to half the distance between the observation points in the width direction to prevent misalignment between the observation points and the center of the cube caused by changing the size of the cubic region.

Next, the number of times to divide the cubic region is determined so that the size of the cubic region is the largest size below the threshold value. By determining the number of times to divide the cubic region in this way, the size of the cubic region increases the further away it is from the sensor (LiDAR2), as shown in Figure 9. In other words, the upper limit of the number of times to divide the octree is determined based on the widthwise spacing of the observation points by the sensor.

Figure 9 is a graph showing the distance between the sensor and the cubic region on the horizontal axis and the size of the cubic region on the vertical axis. The dashed line shows the change in the width direction as 1/2 the shortest distance between observation points. The solid line shows the length of one side of the cubic region. d shows the minimum cubic region size. As will be described later, d is a value determined to prevent the size of the cubic region from becoming too small, and is defined as the length of one side of the cubic region as shown in Figure 9.

If the threshold value for the size of the cubic region becomes smaller than a predetermined minimum size of the cubic region, the number of divisions that results in the predetermined minimum size is used. This involves setting the minimum size of the cubic region in advance, and not dividing it below that size. With the division method described above, the cubic region becomes smaller the closer it is to the LiDAR2, but excessive region division would also be if it becomes too small in relation to the object to be detected. Therefore, excessive region division is suppressed by setting the minimum size in advance.

Returning to the explanation of FIG. 3, the cube processing unit 11a generates a cube region (step S5). In step S5, the cube region is generated by the method described above.

Next, the cube processing unit 11a identifies the relevant cube region (step S6). In step S6, it is determined which of the cube regions that are terminal nodes recursively divided as shown in FIG. 5 each point included in the point cloud information acquired in step S1 is included in. In other words, it is determined which cube region is the terminal node that includes the coordinates indicated by the point within its range.

Next, the cube processing unit 11a increments (+1) the number of observations of the cube region (terminal node) identified in step S2 (step S7). The result of the increment is stored, for example, in the memory of the control unit 11.

Next, the cube processing unit 11a determines whether a predetermined time has elapsed (step S8). If the predetermined time has not elapsed (step S8; NO), the process returns to step S1. Step S8 determines the period for which the number of observations is to be counted, and the predetermined time can be set appropriately to a period during which observation points (point cloud information) can be acquired multiple times depending on the point cloud acquisition interval, etc.

On the other hand, if the predetermined time has elapsed (step S8; YES), the cube processing unit 11a determines whether the number of observations added in step S3 is equal to or greater than a threshold value (step S9). If the number of observations is equal to or greater than a threshold value (observation count threshold value) (step S9; YES), the cube processing unit 11a defines (detects) the cube region currently being focused on as a background region (step S10).

Next, the cube processing unit 11a increments the number of background determinations (step S11). The number of background determinations is the number of times that a region was defined as a background region in step S10.

Next, the cube processing unit 11a determines whether the number of background determinations is equal to or greater than a threshold value (step S12). If the number of background determinations is equal to or greater than a threshold value (background determination threshold value) (step S12; YES), the cube processing unit 11a updates the classification of the cube region defined as the background region in step S10 (step S13). On the other hand, if the number of background determinations is less than the threshold value (step S12; NO), step S13 is not executed and the process proceeds to step S14 described below.

In other words, the cube processing unit 11a functions as a detection unit that detects background regions, which are regions of stationary objects, within a specified space based on point cloud information input to the input unit multiple times over a specified period of time, and detects the background region based on the detection state of the point cloud in each of multiple cube-shaped regions obtained by dividing the specified space using an octree. Furthermore, if the number of times that the point cloud is detected in a cube-shaped region within a specified period of time is equal to or greater than a specified threshold value, the cube processing unit 11a determines that the cube-shaped region is a background region.

Steps S11 to S13 will now be described. When determining the background region using criteria such as those in step S9, some objects, such as parked vehicles and road construction, are temporarily determined to be in the background region, while other objects, such as buildings, are always determined to be in the background region. Therefore, objects determined to be in the background region in step S13 are classified.

For example, as shown in Fig. 10, a plurality of background determination thresholds are set, and objects are classified according to the thresholds. In Fig. 10, classification is made as follows: 0. outside the background, 1. moving objects, 2. stationary objects (short-term remaining objects such as parked vehicles), 3. stationary objects (medium-term remaining objects such as tools and vehicles at a construction site), and 4. stationary objects (long-term remaining objects such as buildings), and thresholds 1, _T12 , _T23 , and _T34 are set as thresholds that are boundaries between each classification. In the explanation of the background determination thresholds, parked vehicles and the like are described, but this is merely an example of classification of the background region, and does not identify objects such as vehicles. Therefore, steps S11 to S13 are different from object detection processing such as clustering, which will be described later.

When the background determination count counter for the cubic region to be determined is 0, the region is classified as 0 (outside background), and when the counter reaches 1, the region is classified as 1 (moving object). Thereafter, when the counter reaches or exceeds threshold values _T12 , _T23 , and _T34 , the region is classified as 1→2, 2→3, and 3→4, respectively. The threshold values _T12 , _T23 , and _T34 are variables and may be set appropriately depending on the installation environment of the object detection device 10. That is, the cubic processing unit 11a includes a counter that counts the number of times a cubic region is detected as a background region, and classifies the background region depending on the number of times the counter counts.

Next, the cube processing unit 11a resets the number of observations and the elapsed time (step S14). The cube processing unit 11a resets the number of observations counted in step S9 and the elapsed time determined in step S8.

On the other hand, if it is determined in step S9 that the number of observations is not equal to or greater than the threshold value (step S9; NO), the cube processing unit 11a defines the currently focused cube region as the detection target region (step S15). The detection target region is a region in which object detection is performed by clustering the point cloud, as described below.

Next, the cube processing unit 11a determines whether the classification of the cube region currently being focused on is other than "0" (step S16). In step S16, it is determined whether the classification is other than "0" among the classifications 0 to 4 shown in FIG. 10, i.e., whether it is outside the background. If the classification is other than "0", the cube processing unit 11a determines whether the region has been determined to be a detection target region for a longer period than the number of times it has been held (step S17). If the region has been determined to be a detection target region for a longer period than the number of times it has been held (step S17; YES), the cube processing unit 11a resets the background determination count counter (step S18). If the region has not been determined to be a detection target region for a longer period than the number of times it has been held (step S17; NO), step S18 is not executed and the process proceeds to step S14.

Steps S16 to S18 will now be described. If the background determination count counter for a cubic area is 1 or greater and the cubic area is determined to be a detection target area, the background determination count counter is generally reset to 0. However, a counter retention count is set for each classification, and the background determination count counter and classification are not reset even if the area is determined to be a detection target area up to that retention count. For example, in the case of parking meters, where parked vehicles are frequently replaced, in the short term it will be classified as a short-term residue of category "2", but in the long term it may be appropriate to treat it as a medium-term residue of category "3". Therefore, by setting the retention count, it is possible to prevent the classification from being reset to "0" when the vehicle moves, etc.

Steps S9 to S18 are executed individually for all cubic regions. In other words, they are repeated as many times as there are cubic regions. For example, if there are 4096 cubic regions, they are repeated 4096 times.

Next, the cube processing unit 11a determines whether there is a parent node having a certain number of child nodes that are background regions (step S19). The certain number in step S19 may be, for example, a majority.

If there is a parent node where a certain number of the child nodes are background regions (step S19; YES), the cube processing unit 11a merges the child nodes having the parent node (step S20). Steps S19 and S20 will be described with reference to FIG. 11. The left side of FIG. 11 is a diagram showing a cube region c that has been determined to be a background region in a specified space S. Also, in the left side of FIG. 11, multiple cube regions c1 are child nodes that have the same parent node, and multiple cube regions c2 are child nodes that have the same parent node.

In this case, if a certain number or more of the child nodes are determined to be background regions, the entire cubic region of the parent node is determined to be the background region. In other words, child nodes other than those determined to be background regions are also determined to be background regions. The right side of Figure 11 shows the parent nodes of cubic regions c1 and c2 as background regions. On the right side of Figure 11, code cb1 indicates the region indicating the parent node of cubic region c1, and code cb2 indicates the region indicating the parent node of cubic region c2. In other words, if a certain number or more of the child nodes connected to the parent node in the octree structure are determined to be background regions, the cube processing unit 11a determines the entire region indicated by the parent node as the background region.

Returning to the explanation of FIG. 3, the detection unit 11b determines whether the cubic region identified in step S5 is a detection target region (step S21). This step S21 is executed after steps up to S20 have been executed and the definition of the background region in the specified space S has been completed at least once. In other words, the observation points (point cloud information) acquired in step S1 are also subjected to the processing from step S21 onwards as a separate process from the processing of steps S7 to S20.

If the identified cubic region is the detection target region (step S21; YES), the detection unit 11b adds the observation point to the object detection processing target (step S22).

The detection unit 11b performs, for example, a clustering process on the observation points that are subject to object detection in step S22, and performs a detection target discrimination process on the generated clusters. The detection target is a process for discriminating the type of moving object, such as a pedestrian or vehicle, from the shape and size of the cluster. When the detection target is discriminated, the detection unit 11b performs, for example, a tracking process on the target and monitors its movement within the specified space S. In other words, the detection unit 11b functions as an object detection unit that detects objects in a specified space based on point cloud information included in areas other than the background area detected by the detection unit.

As is clear from the above explanation, step S1 functions as an input process, steps S2 to S20 function as detection processes, and steps S21 and S22 function as object detection processes.

According to this embodiment, the object detection device 10 is equipped with a cube processing unit 11a that receives point cloud information from a LiDAR 2 that is installed on the ground and scans a scanning range by intermittently irradiating light and receives reflected light of the light to obtain point cloud information based on the reflected light, detects a background region, which is a region of stationary objects, within a specified space S based on the point cloud information input multiple times within a specified time, and a detection unit 11b that detects objects in the specified space S based on point cloud information included in an area other than the background area detected by the cube processing unit 11a. The cube processing unit 11a detects the background region based on the detection state of the point cloud in each of a plurality of cube regions obtained by dividing the specified space S by an octree.

By configuring the object detection device 10 as described above, even if the shape of the background area is complex, it can be approximated by a combination of cubes. Furthermore, by inputting point cloud information, it becomes possible to automatically detect the background area, eliminating the need to set it manually. Furthermore, it becomes possible to automatically update the background area at regular intervals, making it possible to respond to changes in the surrounding environment, such as vegetation, road construction, and snowdrifts.

In addition, the cube processing unit 11a determines that a cube region is a background region when the number of times that a point cloud is detected in a cube region within a specified time period is equal to or greater than a specified threshold. In this way, if a point cloud is detected continuously for a certain period of time, it can be assumed that a stationary object is present.

In addition, when a predetermined number or more of the child nodes connected to a parent node in the octree structure are detected as background regions, the cube processing unit 11a determines the entire area indicated by the parent node as the background region. By doing this, for example, when the majority of the child nodes are determined to be background regions, the area of the parent node is regarded as the background region, and it is not necessary to trace back to the child nodes when identifying the background region. This makes it possible to reference the background region quickly.

The cube processing unit 11a also includes a counter that counts the number of background determinations for the cube region, and classifies the background region according to the number of counts of the counter. In this way, it is possible to change the response when a stationary object is removed for each classification. For example, it is possible to respond quickly to environmental changes for stationary objects observed over a short period of time, and to respond slowly to confirmation for stationary objects observed over a long period of time.

The cube processing unit 11a also sets a polygonal closed area within a specified space S, with a limited range in the horizontal plane and height directions, and detects a background area within the polygonal closed area. By doing this, in a realistic usage environment of the LiDAR2, it is possible to define as the background in advance the excess space in the width and height directions relative to the depth direction. Therefore, the number of cube areas generated can be reduced.

The cube processing unit 11a also divides the cubic region so that the size of the cubic region increases as the distance from the sensor increases within a given space S. In this way, excessive region division is prevented, and appropriate division can be achieved according to the spatial resolution of the LiDAR2.

In addition, the upper limit of the number of divisions using an octree is determined based on the widthwise spacing of the observation points using LiDAR2, which makes it possible to prevent excessive region division.

The present invention is not limited to the above-described embodiment. In other words, a person skilled in the art can implement the present invention by modifying it in various ways in accordance with conventional knowledge without departing from the gist of the present invention. As long as such modifications still include the configuration of the object detection device and object detection method of the present invention, they are of course included in the scope of the present invention.

1 Main body 2 LiDAR (sensor)
REFERENCE SIGNS LIST 10 Object detection device 11 Control unit 11a Cube processing unit 11b Detection unit S Predetermined area C Cube area

Claims (8)

an input unit that receives point cloud information from a sensor that is installed on the ground, intermittently irradiates light to scan a predetermined space, receives reflected light of the light, and acquires point cloud information based on the reflected light;
a detection unit that detects a background region, which is a region of stationary objects, within the specified space based on the point cloud information inputted multiple times to the input unit during a specified time;
an object detection unit that detects an object in the predetermined space based on the point cloud information included in an area other than the background area detected by the detection unit,
the detection unit detects the background region based on a detection state of a point cloud in each of a plurality of cubic regions obtained by dividing the predetermined space by an octree;
1. An object detection device comprising: The object detection device according to claim 1, characterized in that the detection unit determines that the cubic region is the background region when the number of times the point cloud is detected in the cubic region within the specified time period is equal to or greater than a specified threshold. The object detection device according to claim 1, characterized in that, when a predetermined number or more of child nodes connected to a parent node in an octree structure are detected as the background region, the detection unit determines the entire region indicated by the parent node as the background region. the detection unit includes a counter that counts the number of times the cubic region is detected as the background region,
The detection unit classifies the background region according to the count number of the counter.
2. The object detection device according to claim 1 . the detection unit sets an area in the predetermined space, the area being limited in range in a horizontal plane and in a height direction, and detects the background area within the limited area.
2. The object detection device according to claim 1 . The object detection device according to claim 1, characterized in that the detection unit divides the cubic region so that the size of the cubic region increases as the distance from the sensor in the specified space increases. The object detection device according to claim 6, characterized in that the upper limit of the number of divisions by the octree is determined based on the widthwise spacing of the observation points by the sensor. An object detection method implemented by an object detection device having a sensor that is installed on the ground, intermittently irradiates light to scan a predetermined space, receives reflected light of the light, and acquires point cloud information based on the reflected light,
an input step of inputting the point cloud information from the sensor;
a detection step of detecting a background region, which is a region of stationary objects, within the predetermined space based on the point cloud information inputted a plurality of times in the input step during a predetermined time period;
an object detection step of detecting an object in the predetermined space based on the point cloud information included in an area other than the background area detected in the detection step,
The detection step detects the background region based on a detection state of a point cloud in each of a plurality of cubic regions obtained by dividing the predetermined space by an octree.
13. An object detection method comprising:

Images