JP7378302B2 - Object detection method and object detection device - Google Patents

Description

The present invention relates to an object detection method and an object detection device.

Patent Document 1 listed below describes a technique for detecting three-dimensional objects around a vehicle based on point cloud information detected in a detection range around the vehicle.

JP2017-203766A

By acquiring point cloud data that detects the position of the object surface multiple times for the same detection range, it is possible to detect a stationary object that constantly exists at the same position. That is, if a cell is set in the detection range and point cloud data acquired at different times appear in the same cell, it can be determined that the cell is a stationary object cell occupied by a stationary object.

However, even for a cell occupied by a stationary object, there may be cases where point cloud data is not obtained during any of the multiple point cloud data acquisitions. For example, it may be temporarily blocked by another object or may be out of the detection range of the sensor. In such a case, even if a cell is occupied by a stationary object, it may be erroneously determined that it is not a stationary object cell.
The present invention improves the determination accuracy when determining a cell occupied by a stationary object that constantly exists at the same position based on point cloud data of the position of the object surface acquired multiple times for the same detection range. The purpose is to improve.

In the object detection method according to one aspect of the present invention, point cloud data that detects the position of the object surface is acquired multiple times for the same detection range, and the point cloud data is divided into cells for each cell set in the detection range. The cell whose appearance frequency in the point cloud data is equal to or higher than a predetermined frequency threshold is determined to be a stationary object cell, which is a cell occupied by a stationary object that constantly exists at the same location. , if the difference between the frequency at which point cloud data appears in the stationary object cell and the frequency at which point cloud data appears in other cells near the stationary object cell is less than or equal to a predetermined value, the other cells are also stationary. It is determined that it is an object cell.

According to one embodiment of the present invention, a cell occupied by a stationary object that constantly exists at the same position is determined based on point cloud data of the position of the object surface acquired multiple times for the same detection range. The accuracy of judgment can be improved.

FIG. 1 is a schematic configuration diagram of an object detection device according to a first embodiment. FIG. 3 is an explanatory diagram of an example of cell settings. FIG. 7 is a diagram showing the state of the detection range when point cloud data is acquired during the first to fourth and seventh runs. It is a figure which shows the state of the detection range at the time of acquiring point cloud data at the time of the 5th and 6th run. FIG. 7 is a diagram showing an example of the appearance frequency of point cloud data in each cell. FIG. 7 is a diagram showing another example of the appearance frequency of point cloud data in each cell. It is a flow chart of an example of the object detection method of a 1st embodiment. It is a schematic block diagram of the object detection apparatus of 2nd Embodiment. It is a schematic block diagram of the object detection apparatus of 3rd Embodiment. FIG. 6 is an explanatory diagram of the number of times point cloud data can be obtained in each cell, which is the number of times point cloud data can be obtained. It is an explanatory diagram of the number of times that can be obtained. It is a flowchart of an example of the object detection method of 3rd Embodiment.

Embodiments of the present invention will be described below with reference to the drawings. In the description of the drawings, the same or similar parts are given the same or similar symbols, and overlapping explanations are omitted. Each drawing is schematic and may differ from the actual drawing. The embodiments shown below illustrate devices and methods for embodying the technical idea of the present invention, and the technical idea of the present invention is specific to the devices and methods illustrated in the embodiments below. It's not something you do. The technical idea of the present invention can be modified in various ways within the technical scope described in the claims.

(First embodiment)
(composition)
Please refer to FIG. The object detection device 1 includes a first map database 10, a controller 11, and a second map database 12. In FIG. 1, the map database is expressed as "map DB." The same applies to FIGS. 6 and 7.
The first map database 10 is a database of three-dimensional information about roads and their surrounding environment stored in a predetermined storage device such as a flash memory or a hard disk drive.

The three-dimensional information stored in the first map database 10 is collected multiple times by an object sensor such as a camera, a laser radar, a millimeter wave radar, or a distance measurement sensor such as LIDAR (Light Detection and Ranging, Laser Imaging Detection and Ranging). It includes three-dimensional point group data that detects the position of the object surface in the same detection range. Hereinafter, the three-dimensional information obtained by these sensors will be referred to as "point cloud data."

Point cloud data detected (obtained) multiple times in the same detection range can be obtained by, for example, driving a vehicle over a predetermined section multiple times at different times, and detecting the surrounding area using an object sensor mounted on the vehicle. It may be point cloud data obtained by detecting the position of the object surface.
However, it is not necessary to travel with a vehicle equipped with an object sensor, and it is sufficient to acquire point cloud data multiple times at time intervals suitable for determining changes in the position of the object.
In addition to the point cloud data, the first map database 10 contains time information such as the date and time when each point cloud data was acquired, and the number of times the point cloud data was acquired (i.e., how many times the point cloud data was acquired). Information on whether the point cloud data exists or how many times the point cloud data has been acquired is stored in association with the point cloud data.

The controller 11 selects the point cloud data of a stationary object, which is an object that constantly exists at the same position, from among the point cloud data stored in the first map database 10 and stores it in the second map database 12. It is a control unit (ECU: Electronic Control Unit).
The controller 11 includes a processor and peripheral components such as a storage device. The processor may be, for example, a CPU (Central Processing Unit) or an MPU (Micro-Processing Unit).
The storage device may include a semiconductor storage device, a magnetic storage device, an optical storage device, or the like. The storage device may include memory such as a register, a cache memory, a ROM (Read Only Memory) used as a main storage device, and a RAM (Random Access Memory).

The functions of the controller 11 described below are realized, for example, by a processor executing a computer program stored in a storage device.
Note that the controller 11 may be formed of dedicated hardware for executing each information process described below.
For example, the controller 11 may include a functional logic circuit set in a general-purpose semiconductor integrated circuit. For example, the controller 11 may include a programmable logic device (PLD) such as a field-programmable gate array (FPGA).

The second map database 12 stores point cloud data of only stationary objects selected by the controller 11.
The first map database 10 and the second map database 12 may be stored in different storage devices, or may be stored in different storage areas of the same storage device.

Next, an outline of a process in which the controller 11 selects point cloud data of a stationary object from the point cloud data stored in the first map database 10 will be explained.
The controller 11 includes a point cloud reading section 20, a cell setting section 21, a score calculating section 22, and a stationary object determining section 23.
See FIG. 2. The point cloud reading unit 20 reads point cloud data from the first map database 10. When the amount of point cloud data is large, the detection range of the point cloud data to be processed at one time is limited and sequential processing is performed. For example, it is divided into sections 30 of several hundred meters, and processing is performed for each section 30. Hereinafter, the section 30 will be referred to as a "detection range 30."

The point cloud reading unit 20 converts the coordinate system of the read point cloud data into a predetermined processing coordinate system. In this specification, an example will be described in which the processing coordinate system is a three-dimensional Euclidean coordinate system having an X axis, a Y axis, and a Z axis as shown by an arrow 31, but the present invention is not limited thereto. A three-dimensional coordinate system is sufficient as the processing coordinate system.
Note that if the point cloud data stored in the first map database 10 is expressed in advance in a processing coordinate system, there is no need to convert the coordinate system.
Furthermore, all of the point cloud data stored in the first map database 10 may be expressed in the same coordinate system, or may be point cloud data expressed in a sensor coordinate system. In the latter case, the coordinates of the object sensor (coordinates of the vehicle equipped with the object sensor) at the time when the point cloud data is acquired are stored in association with the point cloud data.

The cell setting unit 21 sets a plurality of cells in the detection range 30 based on the point cloud data. For example, the cell setting unit 21 may divide the detection range 30 into a plurality of sections and set cells in the range where point cloud data exists. Although FIG. 2 shows an example in which the detection range 30 is divided into 10×10×10 cells, the number of divisions is not limited to this.
The numbers "1", "2", "3", . . . "9" shown in FIG. 2 indicate the X coordinate, Y coordinate, and Z coordinate of each cell.

If the three-dimensional coordinates of each cell are expressed as (X coordinate, Y coordinate, Z coordinate), the coordinates of cell 32 are (5, 6, 9), and the coordinates of cell 33 are (9, 2, 4). .
As shown in FIG. 2, a cell position expressed by discrete integer coordinate values is sometimes referred to as a "cell position."
Further, in the example of the detection range 30 in FIG. 2, the road surface 34 of the road is included in the cell having the Z coordinate "0".

The score calculation unit 22 calculates a cell score for each cell according to the appearance frequency (appearance rate) of point cloud data appearing in the cell in multiple acquisitions of point cloud data. That is, the score calculation unit 22 calculates the score of the cell according to the number of times point cloud data appears in the cell among the plurality of times point cloud data is acquired. Note that even if a plurality of point cloud data appear in the same cell in the same acquisition time, the number of appearance is determined to be one.

For example, the score calculation unit 22 may calculate the number of appearances itself as the score. Since the number of times point cloud data is acquired is known, the number of appearances represents the frequency of appearance.
The score calculation unit 22 may also calculate the appearance rate (frequency of appearance), which is the ratio of the number of appearances to the number of acquisitions of point cloud data (number of appearances/number of acquisitions), as the score.

Furthermore, the score calculation unit 22 may weight the frequency of appearance according to the interval between acquisition times of the point cloud data, and calculate the score based on the weighted frequency of appearance.
For example, the score calculation unit 22 adds points to the score every time point cloud data appears in a cell, and increases the added points as the time interval between the previous acquisition and the current acquisition is longer.
In this way, by increasing the points added as the time interval between acquisitions becomes longer, the score of a cell occupied by a stationary object increases more easily.

In the following description, an example will be described in which the score calculation unit 22 calculates the number of appearances as a score.
3A and 3B show the detection range 30 viewed from above in the vertical direction (Z-axis direction). Here, a case will be described in which the number of times point cloud data is acquired is seven times.
For example, a vehicle travels through the detection range 30 or a predetermined section around it seven times at different times, and an object sensor mounted on the vehicle detects the position of the object surface around the vehicle to obtain point cloud data. .

Now, as shown in FIG. 3A, during the first to fourth and seventh runs (that is, the first to fourth and seventh point cloud data acquisitions), the road surface 24 in the detection range 30 is When the parked vehicle 35 does not exist and the vehicle runs for the fifth and sixth time (that is, when the point cloud data is acquired for the fifth and sixth time), the parked vehicle 35 temporarily exists as shown in FIG. 3B. Assume that

The parked vehicle 35 exists in the range of X coordinates "1" to "3" and Y coordinates "1" to "5". Further, as shown in FIG. 4A, the Z coordinate of the parked vehicle 35 is "1". FIG. 4A is an XZ plan view of the detection range 30 representing the cell with the Y coordinate "5", and the number in each cell indicates the score. The same applies to FIG. 4B.
During the fifth and sixth acquisitions of point cloud data, the road surface 34 below the parked vehicle 35 is blocked by the parked vehicle 35, so point cloud data that detects the position of the road surface 34 cannot be acquired.

Therefore, as shown in FIG. 4A, the scores (number of appearances) of cells 36, 37, and 38 that include the road surface 34 below the parked vehicle 35 are "5". Further, the scores of cells for other road surfaces 34 are "7". The score of the cell at coordinates (1,5,1) to (3,5,1) occupied by the parked vehicle 35 is "2".
Please refer to FIG. The stationary object determination unit 23 determines whether each cell is occupied by a stationary object, according to the score of the cell calculated by the score calculation unit 22 (that is, the frequency of appearance of point cloud data). Hereinafter, a cell occupied by a stationary object will be referred to as a "stationary object cell."

See FIG. 4A. Since there is a high possibility that point cloud data can always be detected for stationary objects such as road surfaces 34, features such as guardrails, curbs, traffic lights, and signs, and stationary structures such as constructions, stationary object cells occupied by stationary objects The score (frequency of appearance of point cloud data) becomes relatively large.
For example, the scores (number of appearances) of cells 36, 37, and 38 that include the road surface 34 below the parked vehicle 35 are "5". Further, the scores of cells for other road surfaces 34 are "7".

On the other hand, the score of a cell in which no object is present or a cell temporarily occupied by a parked vehicle 35 or a moving object is relatively small.
For example, the score of a cell in which no object exists is "0", and the score of a cell at coordinates (1,5,1) to (3,5,1) temporarily occupied by the parked vehicle 35 is "2". becomes.
For this reason, the stationary object determination unit 23 compares the score of each cell (frequency of appearance of point cloud data) with a frequency threshold set in advance through experiments, etc., and compares the score of each cell (frequency of appearance of point cloud data) with a Cells whose frequency is equal to or higher than a predetermined frequency threshold are determined to be stationary object cells.
Further, the stationary object determination unit 23 determines that a cell with a score less than the frequency threshold is not a stationary object cell. For example, the steady object determination unit 23 determines that a cell with a score greater than 0 and less than the frequency threshold is a cell occupied by an unsteady object (for example, a parked vehicle 35 or a moving object) that temporarily exists in the detection range 30. to decide. In addition, below, a frequency threshold value is also simply described as a threshold value.

As described above, even if the cell is occupied by a stationary object, point cloud data may not be obtained in any of the plurality of point cloud data acquisitions. For example, it may be temporarily blocked by another object or may be out of the detection range of the sensor. Furthermore, when acquiring point cloud data by driving a vehicle as described above, the vehicle does not necessarily travel the same course every time, so the number of measurements for all cells is not necessarily the same.

Therefore, depending on the position of the cell in the XY plane (horizontal plane), the number of times the point cloud data of each cell can be acquired may vary. Therefore, setting a uniform threshold value for different positions in the horizontal plane causes a decrease in the determination accuracy of the stationary object cell.
On the other hand, regarding the vertical direction (Z-axis direction), it is considered that there is little variation in the vertical direction of the object sensor. For example, when an object sensor is mounted on the vehicle mentioned above, the height does not change. Therefore, the number of times point cloud data can be acquired is less dependent on the vertical position of the cell.

For this reason, the steady object determination unit 23 sets a different threshold value for each cell position in the XY plane (horizontal plane). That is, the threshold value is set according to the cell position on the XY plane. In other words, the same threshold value is set for a plurality of cells that have the same horizontal position but different vertical positions.
Hereinafter, the cell position in the XY plane may be referred to as "cell position in the horizontal plane." Further, a cell position in the vertical direction may be referred to as a "vertical cell position".

Here, as shown in FIG. 4A, among a plurality of cells having the same horizontal cell position and different vertical cell positions (for example, cells with coordinates (5,5,0) to (5,5,9)), The score of the stationary object cell tends to be the largest.
For example, among the cells with coordinates (5,5,0) to (5,5,9), the score of the stationary object cell with coordinates (5,5,0) is "7", and the scores of other cells are It is "0". For example, among the cells with coordinates (1,5,0) to (1,5,9), the score of the stationary object cell with coordinates (1,5,0) is "5", and the cell is occupied by the parked vehicle 35. The score of the cell with coordinates (1, 5, 1) is "2", and the scores of the other cells are "0".

For this reason, the stationary object determination unit 23 sets a threshold value according to the maximum score among the scores of a plurality of cells having the same cell position in the horizontal plane and different vertical cell positions. In other words, the threshold value is set according to the highest frequency of occurrence among the frequencies of appearance of point cloud data in a plurality of cells having the same horizontal cell position and different vertical cell positions.
For example, the steady object determination unit 23 may set the threshold value by providing a predetermined margin for the maximum score.

For example, the steady object determination unit 23 sets a value obtained by subtracting "1" from the maximum score as the threshold value. In the example of FIG. 4A, for cells with coordinates (5, 5, 0) to (5, 5, 9), the threshold value "6" is set by subtracting "1" from the maximum score "7". Thereby, the stationary object determining unit 23 determines that the cell with the Z coordinate of “0” and the score of “7” is a stationary object cell.
Furthermore, for cells with coordinates (1, 5, 0) to (1, 5, 9), a threshold value "4" is set, which is the maximum score "5" minus "1". As a result, the stationary object determination unit 23 determines that the cell at the coordinates (1, 5, 0) having the score "5" is a stationary object cell, and the cell at the coordinates (1, 5, 1) having the score "2" It is determined that the cell is not a stationary object cell.

See FIG. 4B. Assume that a stationary object 39 different from the road surface 34 exists at the same cell position in the horizontal plane as the cell position in the horizontal plane where the parked vehicle 35 exists. In the example of FIG. 4B, traffic lights 39 are present in the cell at coordinates (1, 5, 6), the cell at coordinates (2, 5, 6), and the cell at coordinates (3, 5, 6).
In such a case, a difference will occur between the score "7" of the cell occupied by the traffic light 39 and the score "5" of the cell occupied by the road surface 34. In other words, there will be a difference in scores between stationary object cells.

If the threshold value "6" is set here in accordance with the maximum score "7", the cells 36 to 38 occupied by the road surface 34 will be incorrectly determined to be not stationary object cells.
Therefore, in the present invention, it is assumed that other cells near a known stationary object cell (for example, a cell already determined to be a stationary object cell) are also likely to be stationary object cells. This is because stationary objects often exist continuously over a certain range.

If the difference between the score (that is, appearance frequency) of the known stationary object cell 40 and the scores of other cells (for example, cells 36 to 38) near the cell 40 is less than a predetermined value, The other cells (for example, cells 36 to 38) are also determined to be stationary object cells.
For example, the stationary object determination unit 23 determines that the difference between the score "5" of another cell 38 adjacent to the known stationary object cell 40 and the score "7" of the stationary object cell 40 is equal to or less than a predetermined value (for example, "2"). If so, it is determined that the other cells 38 are also stationary object cells.

In addition, the stationary object determination unit 23 determines that the difference between the score "5" of another cell 37 adjacent to the other cell 38 determined to be a stationary object cell and the score "5" of the other cell 38 is a predetermined value. If it is less than a value (for example, "2"), the determination that the other cells 37 are also stationary object cells may be repeated. At this time, instead of the score "5" of the other cell 38, if the difference between the score "7" of the stationary object cell 40 and the score "5" of the other cell 37 is less than a predetermined value, the other cell 37 may also be determined to be a stationary object cell.

Alternatively, the stationary object determination unit 23 compares the scores of cells (for example, cells 36 to 38) in a range a predetermined number (for example, three) away from the known stationary object cell 40 with the score of the stationary object cell 40, Among these cells, a cell whose score is less than or equal to a predetermined value from the score of the stationary object cell 40 may be determined to be a stationary object cell.
The stationary object determination unit 23 stores point cloud data included in cells determined to be stationary object cells in the second map database 12.

(motion)
Next, an example of the object detection method of the first embodiment will be described with reference to FIG.
In step S1, the stationary object determination unit 23 scans the detection range 30 in the X-axis direction and the Y-axis direction, and selects a cell position in the horizontal plane (cell position in the horizontal plane of interest) that is the target of the following steps S2 and S3. do.
In step S2, the stationary object determination unit 23 determines a threshold value according to the maximum score of a plurality of cells having cell positions in the horizontal plane selected in step S1 and having different cell positions in the vertical direction.

In step S3, the stationary object determination unit 23 identifies cells having a score equal to or higher than a threshold value (cells whose appearance frequency in point cloud data is equal to or higher than a predetermined frequency threshold value) among the cells at the cell positions in the horizontal plane selected in step S1 as stationary object determination units. It is determined that it is a cell.
In step S4, the steady object determination unit 23 determines whether there are any remaining cell positions in the horizontal plane that have not yet been selected as cell positions in the horizontal plane of interest. If unselected cell positions in the horizontal plane remain (step S4: N), the process returns to step S1. The steady object determination unit 23 selects any cell position in the horizontal plane that has not yet been selected, and repeats steps S2 and S3. If all the cell positions in the horizontal plane have been selected (step S4: Y), the process advances to step S5.

In step S5, the stationary object determination unit 23 selects any cell that has not been determined to be a stationary object cell (hereinafter sometimes referred to as an "unknown cell"). In the example of FIG. 5, the unknown cell is a cell that was not determined to be a stationary object cell in step S3.
In step S6, the stationary object determination unit 23 calculates the score of the stationary object cell in the vicinity of the unknown cell (frequency of appearance of point cloud data in the stationary object cell) and the score of the unknown cell (frequency of appearance of point cloud data in the unknown cell). It is determined whether the difference is less than or equal to a predetermined value. If the difference in scores is less than or equal to the predetermined value (step S6: Y), the process proceeds to step S7. If the difference in scores is not less than the predetermined value (step S6: N), the process proceeds to step S8. If there is no steady object cell in the vicinity of the unknown cell, the process also proceeds to step S8.

In step S7, the stationary object determination unit 23 determines that the unknown cell is a stationary object cell. Thereafter, the process proceeds to step S9. In step S8, the stationary object determining unit 23 determines that the unknown cell is not a stationary object cell. Thereafter, the process proceeds to step S9.
In step S9, the steady object determination unit 23 determines whether there are any unknown cells remaining that have not been selected in step S5. If unselected unknown cells remain (step S9: N), the process returns to step S5. The steady object determination unit 23 selects any unknown cell that has not been selected yet and repeats steps S6 to S8. If all unknown cells have been selected (step S9: Y), the process ends.

(Effects of the first embodiment)
(1) The first map database 10 stores point cloud data of the position of the object surface obtained multiple times for the same detection range. The score calculation unit 22 obtains the frequency of appearance of point cloud data in each cell set in the detection range. The stationary object determination unit 23 determines the appearance frequency of point cloud data in a stationary object cell, which is a cell whose appearance frequency is equal to or higher than a predetermined frequency threshold, and the appearance frequency of point cloud data in other cells in the vicinity of the stationary object cell. When the difference between the appearance frequency of the group data and the appearance frequency is less than or equal to a predetermined value, other cells are also determined to be stationary object cells.
This can reduce the possibility that a cell temporarily blocked by another object is incorrectly determined as not a stationary object cell when acquiring point cloud data. Thereby, the determination accuracy when determining a stationary object cell can be improved.

(2) If the frequency of occurrence of point cloud data in other cells is not zero, and the difference from the frequency of occurrence of point cloud data in a steady object cell is larger than a predetermined value, It may be determined that another cell is occupied by an unsteady object that temporarily exists in the detection range.
Thereby, cells occupied by non-stationary objects can be detected.

(3) Point cloud data acquired multiple times is obtained by driving a vehicle equipped with a sensor that detects the position of the object surface multiple times in or around the detection range, and using the sensor to detect the position of the object surface within the detection range. It may be point cloud data obtained by
As a result, point cloud data can be acquired at different times with appropriate intervals, so that if an object appears in the same cell acquired at these times, it can be determined that it is a stationary object cell.

(4) The stationary object determination unit 23 determines that the stationary object cell and another cell are adjacent to each other, and the difference between the frequency of appearance of point cloud data in the other cell and the frequency of appearance of point cloud data in the stationary object cell is a predetermined value. If it is less than or equal to the value, the other cell may be determined to be a stationary object cell.
This can reduce the possibility that a cell temporarily blocked by another object is incorrectly determined as not a stationary object cell when acquiring point cloud data. Thereby, the determination accuracy when determining a stationary object cell can be improved.

(5) The steady object determination unit 23 sets the same frequency threshold for a plurality of cells that have the same horizontal position and different vertical positions. The steady object determination unit 23 determines a frequency threshold based on the frequency of appearance of point cloud data in a plurality of cells at different vertical positions.
As a result, even if the number of times point cloud data can be acquired for each cell varies depending on the cell position in the horizontal plane, it is possible to set a frequency threshold according to the cell position in the horizontal plane, improving judgment accuracy. .

(6) The stationary object determination unit 23 determines the frequency of appearance of point cloud data in a second other cell adjacent to another cell determined to be a stationary object cell, and the point group data in a stationary object cell adjacent to another cell. If the difference between the appearance frequency of group data or the appearance frequency of point group data in another cell is less than or equal to a predetermined value, it is determined that the second other cell is also a stationary object cell.
Thereby, even if a plurality of cells are temporarily blocked by other objects when acquiring point cloud data, it is possible to reduce the possibility of erroneously determining that the cell is not a stationary object cell. Thereby, the determination accuracy when determining a stationary object cell can be improved.

(7) The score calculation unit 22 calculates a score by weighting the frequency of appearance of point cloud data in a cell according to the interval of acquisition time of the point cloud data. The stationary object determination unit 23 determines whether the cell is a stationary object cell based on a score that is a weighted appearance frequency.
The longer the interval between acquisition times of point cloud data acquired in the same cell, the higher the possibility that this cell will be occupied by an object for a long time, so it can be determined that there is a higher possibility that a stationary object exists. Therefore, the accuracy of the determination can be improved by making the determination based on the appearance frequency weighted according to the interval of acquisition times.

(Second embodiment)
Next, a second embodiment will be described. In the first embodiment, constantly existing objects such as the road surface 34, features such as guardrails, curbs, traffic lights, and signs, and stationary structures such as constructions are detected as stationary objects, and objects such as the parked vehicle 35 are detected as stationary objects. Stationary objects and moving objects that remain stationary for a relatively short period of time are detected as unsteady objects.
In the second embodiment, moving objects are detected as non-stationary objects, and stationary objects are detected as stationary objects. The object detection device 1 of the second embodiment can be used to distinguish point cloud data acquired by an object sensor into point cloud data of a moving object and point cloud data of a stationary object.

FIG. 6 is a schematic configuration diagram of an object detection device 1 according to the second embodiment. The object detection device 1 of the second embodiment includes an object sensor 13 and a positioning device 14.
The object sensor 13 includes a plurality of different types of sensors that detect the position of the object surface around the vehicle (hereinafter referred to as "own vehicle") in which the object detection device 1 is mounted and output it as point cloud data. Be prepared. The object sensor 13 may be, for example, a camera, a distance measuring sensor such as a laser radar, a millimeter wave radar, or a LIDAR (Light Detection and Ranging, Laser Imaging Detection and Ranging).

The positioning device 14 measures the current position of the own vehicle. The positioning device 14 may measure the current position of the own vehicle using, for example, a global positioning system or odometry. Furthermore, the positioning device 14 may measure the current position of the own vehicle using map matching between the point cloud data acquired by the object sensor 13 and known point cloud data.
The point cloud reading unit 20 reads point cloud data acquired by the object sensor 13 at different times in the same detection range as point cloud data detected (obtained) multiple times. The point cloud reading unit 20 converts the coordinate system of the point cloud data into a predetermined processing coordinate system based on the current position of the own vehicle.

The difference between the point cloud data acquired multiple times in the first embodiment and the point cloud data acquired multiple times in the second embodiment is the length of the acquisition interval at which point cloud data is acquired multiple times. The interval between acquisition times in the first embodiment is relatively long (for example, several tens of minutes to several days), whereas the interval between acquisition times in the second embodiment is relatively short (for example, several tens of milliseconds to several seconds). .
For example, the object sensor 13 may detect and output point cloud data from a captured image for each frame of a video captured by a camera. In this case, the interval between point group data acquisition times corresponds to the frame rate of the captured video.
For example, the object sensor 13 may output point cloud data for each scan cycle of laser radar, millimeter wave radar, or LIDAR. In this case, the interval between acquisition times is a scan cycle.

The stationary object determination unit 23 detects stationary object cells and cells occupied by unsteady objects, as in the first embodiment. Based on the point cloud data acquired multiple times at relatively short intervals as described above, when stationary object cells and cells occupied by unsteady objects are detected, cells occupied by moving objects are occupied by unsteady objects. A cell occupied by a stationary object can be detected as a stationary object cell.
The stationary object determination unit 23 removes point cloud data included in cells occupied by moving objects from the point cloud data acquired from the object sensor 13 and stores the data in the second map database 12 .

(Effects of the second embodiment)
The stationary object determination unit 23 can detect cells occupied by moving objects as cells occupied by unsteady objects, and can detect cells occupied by stationary objects as stationary object cells.
Thereby, it can be determined whether the point cloud data acquired from the object sensor 13 is point cloud data that detected a moving object or point cloud data that detected a stationary object.

(Third embodiment)
Next, a third embodiment will be described. As mentioned above, depending on the cell position in the horizontal plane, there may be unevenness in the number of times that point cloud data can be acquired for each cell. Compared to the score, this causes a decrease in the determination accuracy of the stationary object cell.
Therefore, in the first embodiment, a threshold value is set for each cell position in the horizontal plane, and the threshold value is determined based on the scores of cells at different vertical cell positions within the same cell position in the horizontal plane.

On the other hand, in the third embodiment, a threshold value is set for each cell depending on the number of times each cell can be obtained.
FIG. 7 is a schematic configuration diagram of an object detection device 1 according to the third embodiment. The object detection device 1 of the third embodiment includes a storage device that stores vehicle position data 15.
The vehicle position data 15 is position data indicating the position of a vehicle equipped with an object sensor that acquires the point cloud data stored in the first map database 10 at the time of acquiring the point cloud data.
The first map database 10 and the second map database 12 may be stored in different storage devices, or may be stored in different storage areas of the same storage device.

Based on the point cloud data read from the first map database 10 by the point cloud reading section 20 and the vehicle position data 15, the stationary object determination section 23 performs the following for each cell in acquiring point cloud data multiple times. Determine the possible number of acquisitions, which is the number of times point cloud data can be acquired in the cell.
See FIGS. 8A and 8B. For example, assume that a vehicle 50 equipped with an object sensor is driven seven times and point cloud data for seven times is acquired.

Here, the parked vehicle 53 temporarily exists in the detection range 30 during the first to third runs (that is, when the first to third point cloud data are acquired), and the parked vehicle 53 temporarily exists in the detection range 30 during the fourth to seventh runs. Let us consider a case where the parked vehicle 53 has already moved when traveling (that is, when the point cloud data is acquired for the fourth to seventh times).
At this time, since the cell 51 is not blocked by the parked vehicle 53, the steady object determination unit 23 determines that the number of times the cell 51 can be acquired is "7".

On the other hand, during the first to third runs, the cell 52 is blocked by the parked vehicle 53, and point cloud data within the cell 52 cannot be acquired. Therefore, the steady object determination unit 23 determines that the number of times the cell 52 can be acquired is "4".
The steady object determination unit 23 determines whether or not the cell is blocked by the object indicated by the point cloud data, based on the position of the vehicle 50 indicated by the vehicle position data 15, the position of the object indicated by the point cloud data, and the position of the cell. By determining whether point cloud data can be acquired in the cell, it is determined whether or not point cloud data can be acquired in the cell.

Then, the steady object determination unit 23 sets a threshold value according to the number of times each cell can be acquired. The steady object determination unit 23 may set a threshold value by providing a predetermined margin for the number of times the object can be acquired.
For example, the steady object determination unit 23 may set a value obtained by subtracting "1" from the number of times the object can be obtained as the threshold value. For example, the threshold value of the cell 51 where the number of possible acquisitions is "7" may be set to "6", and the threshold value of the cell 52 where the number of possible acquisitions is "4" may be set to "3".

The stationary object determination unit 23 may calculate the threshold value by weighting the number of times the point cloud data can be acquired according to the interval between acquisition times of the point cloud data, in the same way as calculating the score. For example, the stationary object determination unit 23 adds points to the threshold value each time an acquisition time in which point cloud data can be acquired arrives, and increases the addition points as the time interval between the previous acquisition time and the current acquisition time becomes longer. good.
The stationary object determination unit 23 compares the score of the cell with a threshold value set for each cell, and determines that the cell is a stationary object cell if the score is equal to or greater than the threshold value.

(motion)
An example of the object detection method of the third embodiment will be described with reference to FIG.
In step S11, the stationary object determination unit 23 selects any cell within the detection range 30.
In step S12, the steady object determination unit 23 determines the number of times the cell selected in step S11 can be obtained, and determines a threshold value according to the number of times the cell can be obtained.

In step S13, the stationary object determination unit 23 determines that the cell selected in step S11 is a stationary object cell if the score of the cell selected in step S11 is greater than or equal to the threshold value.
In step S14, the stationary object determination unit 23 determines whether there are any remaining cells that have not been selected in step S11. If unselected cells remain (step S14: N), the process returns to step S11. The steady object determination unit 23 selects any cell that has not been selected yet and repeats steps S12 and S13. If all cells have been selected (step S4: Y), the process advances to step S15.
The processing in steps S15 to S19 is similar to the processing in steps S5 to S9 in FIG.

(Effects of the third embodiment)
The steady object determination unit 23 determines, for each cell, the number of times point cloud data can be acquired, which is the number of times point cloud data can be acquired in each cell, out of the plurality of times point cloud data is acquired. The stationary object determination unit 23 sets a threshold value for each cell according to the number of times it can be acquired, and determines a cell in which the frequency of appearance of point cloud data is equal to or greater than the threshold value to be a stationary object cell.
By setting the threshold value for each cell according to the number of times it can be acquired in this way, even if the number of times point cloud data can be acquired for each cell is uneven, the threshold value can be set according to the number of times it can be acquired. This makes it possible to improve the judgment accuracy.

DESCRIPTION OF SYMBOLS 1... Object detection device, 10... First map database, 11... Controller, 12... Second map database, 13... Object sensor, 14... Positioning device, 15... Vehicle position data, 20... Point cloud reading unit, 21... Cell Setting section, 22... Score calculation section, 23... Stationary object determination section

Claims (10)

Point cloud data obtained by detecting the position of the object surface with a sensor is acquired multiple times for the same detection range,
For each cell set in the detection range, obtain the frequency at which the point cloud data appears in the cell,
A cell in which the appearance frequency of the point cloud data is equal to or higher than a predetermined frequency threshold is determined to be a stationary object cell, which is a cell occupied by a stationary object that constantly exists at the same position, and the stationary object cell is If the difference between the frequency of appearance of the point cloud data and the frequency of appearance of the point cloud data in other cells near the stationary object cell is less than or equal to a predetermined value, the other cell is also a stationary object cell. It is determined that there is
Setting the same frequency threshold for a plurality of cells having the same horizontal position and different vertical positions,
determining the frequency threshold based on the frequency at which the point cloud data appears in the plurality of cells at different vertical positions;
An object detection method characterized by: Point cloud data obtained by detecting the position of the object surface with a sensor is acquired multiple times for the same detection range,
For each cell set in the detection range, obtain the frequency at which the point cloud data appears in the cell,
A cell in which the appearance frequency of the point cloud data is equal to or higher than a predetermined frequency threshold is determined to be a stationary object cell, which is a cell occupied by a stationary object that constantly exists at the same position, and the stationary object cell is If the difference between the frequency of appearance of the point cloud data and the frequency of appearance of the point cloud data in other cells near the stationary object cell is less than or equal to a predetermined value, the other cell is also a stationary object cell. It is determined that there is
weighting the frequency of appearance of the point cloud data in the cell according to the interval of acquisition time of the point cloud data, and determining whether the cell is a stationary object cell based on the weighted frequency of appearance;
An object detection method characterized by: If the frequency of appearance of the point cloud data in the other cell is not zero, and the difference between the frequency of occurrence of the point cloud data in the stationary object cell is greater than the predetermined value, the other cell is 3. The object detection method according to claim 1 , further comprising determining that the cell is occupied by an unsteady object that temporarily exists in the detection range. The point cloud data acquired multiple times includes a vehicle equipped with the sensor that detects the position of an object surface that travels in or around the detection range multiple times, and the sensor detects the position of the object surface within the detection range. 4. The object detection method according to claim 3 , wherein the point group data is obtained by detecting. 4. The object detection method according to claim 3 , wherein the stationary object is a stationary object, and the unsteady object is a moving object. The stationary object cell and the other cell are adjacent to each other, and the difference between the frequency at which the point cloud data appears in the other cell and the frequency at which the point cloud data appears in the stationary object cell is equal to or less than the predetermined value. The object detection method according to any one of claims 1 to 5 , characterized in that, in certain cases, the other cell is determined to be a stationary object cell. Determining, for each cell, a possible number of acquisitions, which is the number of times the point cloud data can be acquired in each cell, among the plurality of times of acquiring the point cloud data;
setting the frequency threshold for each cell according to the number of times it can be acquired;
The object detection method according to any one of claims 1 to 6 , characterized in that: The frequency of appearance of the point cloud data in a second other cell adjacent to the other cell determined to be a stationary object cell, and the occurrence of the point cloud data in the stationary object cell adjacent to the other cell. The second other cell is also determined to be a stationary object cell when the difference between the frequency or the frequency of appearance of the point cloud data in the other cell is less than or equal to the predetermined value. The object detection method according to any one of claims 1 to 7. a storage device that stores point cloud data of the position of the object surface obtained multiple times by detecting the position of the object surface with a sensor in the same detection range;
For each cell set in the detection range, the frequency at which the point cloud data appears in the cell is acquired, and cells whose frequency of appearance of the point cloud data is equal to or higher than a predetermined frequency threshold are constantly located at the same location. The cell is determined to be a stationary object cell , which is a cell occupied by a stationary object existing in a controller that determines that the other cell is also a stationary object cell when the difference between the frequency of appearance and the cell is less than or equal to a predetermined value;
Equipped with
The controller sets the same frequency threshold for a plurality of cells that have the same horizontal position and different vertical positions, and sets the frequency threshold for each of the cells that have different vertical positions. An object detection device characterized in that the frequency threshold is determined based on the frequency threshold value . a storage device that stores point cloud data of the position of the object surface obtained multiple times by detecting the position of the object surface with a sensor in the same detection range;
For each cell set in the detection range, the frequency at which the point cloud data appears in the cell is acquired, and cells whose frequency of appearance of the point cloud data is equal to or higher than a predetermined frequency threshold are constantly located at the same location. The cell is determined to be a stationary object cell, which is a cell occupied by a stationary object existing in a controller that determines that the other cell is also a stationary object cell when the difference between the frequency of appearance and the cell is less than or equal to a predetermined value;
Equipped with
The controller weights the frequency of occurrence of the point cloud data in the cell according to the interval between acquisition times of the point cloud data, and determines whether the cell is a stationary object cell based on the weighted frequency of appearance. An object detection device characterized by:

Images