JP7149172B2 - Object recognition method and object recognition device - Google Patents

Description

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

Conventionally, an object tracking method using a laser radar is known (Patent Document 1). The invention described in Patent Document 1 creates a cluster using detection points detected by a laser radar, calculates the matching degree of the shape of the cluster based on the amount of movement of the detection points included in the cluster, and calculates the degree of matching. A cluster is specified as one object when the matching degree of the shape obtained is equal to or higher than a predetermined threshold. As a result, the invention described in Patent Document 1 can identify an object and also calculate the velocity of the object.

JP 2013-228259 A

Since an object coming out of the blind spot of the sensor gradually becomes visible, the appearance of the object as seen from the own vehicle changes greatly. However, the invention described in Patent Literature 1 does not refer to an object coming out of the blind spot of the sensor, and there is a possibility that the speed of the object coming out of the blind spot of the sensor cannot be accurately calculated.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problem, and an object thereof is to provide an object recognition method and an object recognition apparatus capable of accurately calculating the velocity of an object.

An object recognition method according to an aspect of the present invention calculates a position of a blind spot boundary indicating a boundary between a sensor detectable range and a sensor undetectable range. The object recognition method acquires a map database using a sensor and acquires the position of the own vehicle. The object recognition method determines whether or not the vehicle is approaching a junction based on the map database and the position of the vehicle. The object recognition method sets a blind spot boundary on the second lane that merges with the first lane on which the own vehicle is traveling when the own vehicle is approaching a merging point. In the object recognition method, when the own vehicle is approaching a merging point and an object is recognized between the first lane and the second lane, the object is judged as an obstacle, and clustered edges of the obstacle are identified. A blind spot boundary is set on an extension of a straight line connecting the detection point of the part and the sensor. The object recognition method clusters a plurality of detection points detected by a sensor, and determines whether the clustered cluster touches the blind spot boundary. The object recognition method extracts a detection point farthest from the blind spot boundary from among a plurality of detection points included in a cluster that touches the blind spot boundary, and tracks the extracted detection point.

According to the present invention, it is possible to accurately calculate the velocity of an object.

FIG. 1 is a schematic configuration diagram of an object recognition device according to the first embodiment of the present invention. FIG. 2 is a flow chart explaining an operation example of the object recognition device according to the first embodiment of the present invention. FIG. 3 is a diagram illustrating an example of a blind spot boundary. FIG. 4 is a schematic configuration diagram of an object recognition device according to a second embodiment of the present invention. FIG. 5 is a flow chart illustrating an operation example of the object recognition device according to the second embodiment of the present invention. FIG. 6 is a diagram illustrating an example of a blind spot boundary setting method. FIG. 7 is a diagram explaining an example of a tracking method.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are denoted by the same reference numerals, and the description thereof is omitted.

(First embodiment)
(Configuration example of object recognition device 1)
A configuration example of the object recognition device 1 will be described with reference to FIG. As shown in FIG. 1, the object recognition device 1 includes a sensor 10, a controller 20, and a display 30.

The object recognition device 1 may be installed in a vehicle having an automatic driving function, or may be installed in a vehicle without an automatic driving function. Further, the object recognition device 1 may be installed in a vehicle capable of switching between automatic driving and manual driving. Note that automatic driving in this embodiment refers to, for example, a state in which at least one of actuators such as a brake, an accelerator, and a steering is controlled without being operated by a passenger. Therefore, other actuators may be operated by the passenger's operation. Further, automatic operation may be a state in which any control such as acceleration/deceleration control or lateral position control is being executed. Further, manual driving in this embodiment refers to a state in which the driver is operating the brake, accelerator, and steering, for example.

The sensor 10 is a device that is mounted on the own vehicle and detects objects around the own vehicle. The sensor 10 includes lidar, radar, millimeter wave radar, laser range finder, sonar, and the like. The sensor 10 detects moving objects including other vehicles, motorcycles, bicycles and pedestrians, and stationary objects including obstacles, falling objects and parked vehicles as objects around the own vehicle. The sensor 10 also detects the position, orientation (yaw angle), size, speed, acceleration, deceleration, and yaw rate of the moving and stationary objects with respect to the host vehicle. Also, the sensor 10 may include a wheel speed sensor, a steering angle sensor, a gyro sensor, and the like. The sensor 10 outputs detected information to the controller 20 .

In this embodiment, sensor 10 is described as lidar or radar. A lidar or radar scans an object with radio waves and measures the reflected wave to measure the distance and direction to the object. Lidar or radar also acquires point clouds of objects as detection points. A point cloud is a collection of points processed by the controller 20 . A point cloud is usually represented by three-dimensional coordinates (x, y, z). Note that the method of acquiring the point cloud of an object is not limited to lidar or radar. For example, there is a method of obtaining a point cloud of pixels each having a feature distinguishable from surrounding pixels from a camera image. This method includes, for example, the non-patent document "Jianbo Shi and Carlo Tomasi, "Good Features to Track," 1994 IEEE Conference on Computer Vision and Pattern Recognition (CVPR'94), 1994, pp.593-600." method is used.

The controller 20 is a general-purpose microcomputer having a CPU (Central Processing Unit), memory, and an input/output unit. A computer program for functioning as the object recognition device 1 is installed in the microcomputer. By executing the computer program, the microcomputer functions as a plurality of information processing circuits included in the object recognition device 1. FIG. Here, an example of realizing a plurality of information processing circuits provided in the object recognition apparatus 1 by software will be shown. It is also possible to construct a circuit. Also, a plurality of information processing circuits may be configured by individual hardware. The controller 20 includes a blind spot boundary calculation unit 21, an object recognition unit 22, and a driving support unit 23 as a plurality of information processing circuits.

The blind spot boundary calculator 21 calculates the position of the blind spot boundary of the sensor 10 and sets the blind spot boundary based on the calculation result. The object recognition unit 22 clusters and tracks detection points detected by the sensor 10 . The driving assistance unit 23 performs driving assistance based on the processing result of the object recognition unit 22 .

Next, an operation example of the object recognition device 1 will be described with reference to FIGS. 2 and 3. FIG.

In step S101 shown in FIG. 2, the blind spot boundary calculator 21 calculates the position of the blind spot boundary of the sensor 10, and sets the blind spot boundary based on the calculation result. In the present embodiment, the blind spot boundary indicates a boundary between a range detectable by the sensor 10 and a range undetectable by the sensor 10 . Normally, the sensor 10 is set with a detectable range (detection distance) as an indicator of performance. The blind spot boundary calculator 21 calculates the position of the blind spot boundary of the sensor 10 based on the performance (detection distance) of the sensor 10 . As an example, the blind spot boundary is indicated by 51 in FIG. A region 50 (hatched area in FIG. 3) surrounded by a blind spot boundary 51 indicates a detectable range of the sensor 10 . In addition, in FIG. 3, the shape of the blind spot boundary 51 is a semicircular shape, but is not limited to this. The shape of the blind spot boundary 51 includes a circular shape, a fan shape, a rectangular shape, and a polygonal shape. The calculation method (setting method) of the blind spot boundary 51 is not limited to this. Other methods are described in the second embodiment.

The process proceeds to step 103, and the sensor 10 detects objects around the host vehicle 40 (other vehicles 60 and 61 shown in FIG. 3). As shown in FIG. 3, the sensor 10 detects a plurality of detection points (detection points 70 to 77) on the surfaces of other vehicles 60 and 61. As shown in FIG.

The process proceeds to step 105 , and the object recognition unit 22 clusters the detection points 70 to 77 detected in step 103 . Clustering is a method of extracting a plurality of detection points as clusters (objects). As an example, the object recognition unit 22 clusters detection points whose distance between two points is equal to or less than a predetermined value from a plurality of detection points. Specifically, the detection points 70 to 75 shown in FIG. 3 are clustered and extracted as one cluster 60 (another vehicle 60). Similarly, the detection points 76 to 77 shown in FIG. 3 are clustered and extracted as one cluster 61 (another vehicle 61).

The process proceeds to step 107 , and the object recognition unit 22 determines whether the clusters 60 and 61 extracted in step S 105 are in contact with the blind spot boundary 51 . In the example shown in FIG. 3, the cluster 61 abuts the blind spot boundary 51 . On the other hand, cluster 60 is not in contact with blind spot boundary 51 . In this case, the object recognition unit 22 determines that the cluster 61 contacts the blind spot boundary 51 . Also, the object recognition unit 22 determines that the cluster 60 is not in contact with the blind spot boundary 51 .

A cluster 61 bordering the blind spot boundary 51 is an object coming out of the blind spot of the sensor 10 or disappearing into the blind spot of the sensor 10 . In this embodiment, the cluster 61 that contacts the blind spot boundary 51 will be described as an object coming out of the blind spot of the sensor 10 . Since the object (cluster 61) emerging from the blind spot of the sensor 10 gradually becomes visible, the appearance of the object seen from the sensor 10 changes greatly.

Therefore, the object recognition unit 22 extracts the detection point (detection point 77) farthest from the blind spot boundary 51 from a plurality of detection points (detection points 76 to 77) included in the cluster 61 that contacts the blind spot boundary 51. (Step S109). The process proceeds to step S111, and the object recognition unit 22 tracks the detection point 77 extracted in step S109. In this way, the object recognition unit 22 tracks the detection point (detection point 77) that is the farthest from the blind spot boundary 51, so that even in a scene where the appearance of the object (other vehicle 61) seen from the sensor 10 changes greatly, , the object recognition unit 22 can accurately calculate the speed of the object (another vehicle 61).

A plurality of detection points (detection points 70 to 75) included in the cluster 60 not in contact with the blind spot boundary 51 are tracked by a well-known method (No in step S107).

The process proceeds to step 113, and the driving support unit 23 displays information on the other vehicles 60 and 61 tracked in step S111 on the display 30 to support driving.

The process proceeds to step 115, and when the occupant of the own vehicle 40 turns off the ignition (Yes in step S115), the series of processes ends. If the user has not turned off the ignition (No in step S115), the process returns to step S103. Turning off the ignition in this embodiment includes stopping the vehicle 40 and stopping the power supply system of the vehicle 40 . Turning off the ignition may be realized by turning off an ignition switch provided in the vehicle interior of the vehicle 40, or may be realized by turning off a power system switch provided in the vehicle interior of the vehicle 40. good.

(Effect)
As described above, according to the object recognition device 1 according to the first embodiment, the following effects are obtained.

The object recognition device 1 calculates the position of a blind spot boundary 51 that indicates the boundary between the range detectable by the sensor 10 and the range undetectable by the sensor 10, and sets the blind spot boundary based on the calculation result. The object recognition device 1 clusters a plurality of detection points detected by the sensor 10 (detection points 70 to 75 and 76 to 77 shown in FIG. 3). The object recognition device 1 determines whether or not the clustered clusters (the clusters 60 and 61 shown in FIG. 3) touch the blind spot boundary 51 . The object recognition device 1 extracts a detection point 77 farthest from the blind spot boundary 51 from among a plurality of detection points (detection points 76 to 77 shown in FIG. 3) included in the cluster 61 contacting the blind spot boundary 51 . Then, the object recognition device 1 tracks the extracted detection point 77 . As a result, the object recognition apparatus 1 can accurately calculate the speed of the other vehicle 61 even in a scene where the appearance of the object (the other vehicle 61 shown in FIG. 3) as seen from the sensor 10 changes significantly.

Also, since the blind spot boundary 51 is set, the object recognition device 1 can quickly recognize an object that has entered the detectable range of the sensor 10 .

Also, the object recognition device 1 tracks the detection points 76 and 77 after clustering them. As a result, noise unrelated to the running of the own vehicle 40 is removed.

(Second embodiment)
(Configuration example of object recognition device 2)
Next, a second embodiment of the present invention will be described with reference to FIGS. 4 to 7. FIG. Reference numerals are used for configurations that overlap with the first embodiment, and descriptions thereof are omitted. The following description will focus on the differences.

As shown in FIG. 4, the object recognition device 2 according to the second embodiment further includes a map database 11 and a GPS receiver 12. FIG.

The map database 11 is a database stored in a car navigation device or the like, and stores various data necessary for route guidance such as road information and facility information. The map database 11 also stores information such as the structure of roads, the number of lanes on roads, the positions of lanes, road boundaries, and targets. The map database 11 outputs map information to the controller 20 in response to requests from the controller 20 . Various data such as road information and target information are not necessarily acquired from the map database 11, and may be acquired by the sensor 10, or may be acquired using vehicle-to-vehicle communication or road-to-vehicle communication. good too. In addition, when various data such as road information and target information are stored in a server installed outside, the controller 20 may acquire these data from the server at any time through communication. Further, the controller 20 may periodically acquire the latest map information from an externally installed server and update the map information it holds.

The GPS receiver 12 detects the position of the vehicle 40 on the ground (hereinafter sometimes referred to as the self-position) and the attitude of the vehicle 40 by receiving radio waves from satellites. The GPS receiver 12 outputs the detected position information and attitude information of the own vehicle 40 to the controller 20 .

Next, an operation example of the object recognition device 2 will be described with reference to FIGS. 5 to 7. FIG.

At step 201 , the sensor 10 detects objects around the host vehicle 40 .

The process proceeds to step 203 and the GPS receiver 12 detects the self position and the attitude of the own vehicle 40 .

The process proceeds to step 205 , and the object recognition section 22 acquires map information around the self-position from the map database 11 .

The process proceeds to step 207, and the object recognition unit 22 determines that the vehicle 40 is running based on the self-position and attitude of the vehicle 40 detected in step S203 and the map information acquired in step S205. get the lane In addition, below, the lane in which the own vehicle 40 is driving is simply called the own lane. 90 shown in FIG. 6 is the own lane. Further, the object recognition unit 22 determines whether or not the vehicle 40 is approaching a merging point or an intersection based on the self-position and attitude of the vehicle 40 detected in step S203 and the map information acquired in step S205. determine whether As shown in FIG. 6, when the vehicle 40 is approaching the junction (Yes in step S207), the process proceeds to step S209. On the other hand, if the host vehicle 40 is not approaching the junction or intersection (No in step S207), the process proceeds to step S221.

In step S209, the object recognition unit 22 acquires information (positional information, structure information) of the lane 91 that merges with the own lane 90 from the map database 11 (see FIG. 6). Although illustration is omitted, when the own vehicle 40 is approaching an intersection, the object recognition unit 22 acquires information on lanes intersecting with the own lane from the map database 11 .

The process proceeds to step S211, and the blind spot boundary calculator 21 determines whether or not the sensor 10 has a blind spot. Specifically, blind spot boundary calculator 21 determines whether or not a plurality of detection points are detected between own lane 90 and lane 91 . As shown in FIG. 6, when a plurality of detection points 80-87 are detected between the own lane 90 and the lane 91, the detection points 80-87 are clustered and recognized as an object. Further, this object is determined as an obstacle 92 (eg, a wall). Whether or not an object is an obstacle is determined, for example, based on map information. As shown in FIG. 6, when there is a blind spot of the sensor 10 (Yes in step S211), the process proceeds to step S213. On the other hand, if there is no blind spot of the sensor 10 (No in step S211), the process proceeds to step S221.

In step S213, the blind spot boundary calculator 21 sets the blind spot boundary on a straight extension line 93 connecting the own vehicle 40 (sensor 10) and the detection points 80 at the ends of the clustered obstacles 92 (Fig. 6). In FIG. 6, the other vehicle 62 is hidden in a blind spot and is not detected, but in FIG. 7, detection points 88 and 89 of the other vehicle 62 are detected.

The process proceeds to step 215, and the object recognition section 22 clusters the detection points 88 and 89 shown in FIG. The process proceeds to step 217, and the object recognition unit 22 determines whether or not the clustered cluster (other vehicle 62) is in contact with the blind spot boundary. As shown in FIG. 7, when the cluster (other vehicle 62) is in contact with the blind spot boundary (Yes in step S217), the process proceeds to step S219.

In step S219, the object recognition unit 22 extracts the detection point (detection point 89 shown in FIG. 7) that is farthest from the blind spot boundary. Detection point 89 is the farthest detection point from the blind spot boundary on lane 91 . The process proceeds to step S221, and the object recognition unit 22 tracks the detection point 89 extracted in step S219.

The process proceeds to step 223, and the object recognition section 22 calculates the velocity vector of the detection point 89 based on the tracking result. As an example, the object recognition unit 22 calculates the velocity vector of the detection point 89 from the temporal change in the position of the detection point 89 . Then, the object recognition unit 22 decomposes the velocity vector into a velocity component in the traveling direction of the lane 91 and a velocity component in a direction perpendicular to the traveling direction.

The process proceeds to step 225 , and the driving support unit 23 uses the speed component in the traveling direction of the lane 91 calculated in step 223 to calculate the time required for the other vehicle 62 to reach the merging point. Then, the driving support unit 23 displays the calculated time on the display 30 to perform driving support.

(Effect)
As described above, according to the object recognition device 2 according to the second embodiment, the following effects are obtained.

The object recognition device 2 determines whether or not the vehicle 40 is approaching a merging point or an intersection based on the self-position and attitude of the vehicle 40 detected by the GPS receiver 12 and the map information acquired from the map database 11. determine whether When the own vehicle 40 is approaching a merging point or an intersection, the object recognition device 2 sets a blind spot boundary on the lane 91 that joins or intersects with the own lane 90 (first lane). As a result, the object recognition device 2 can quickly recognize an object even at a junction or an intersection with poor visibility, and can accurately calculate the speed of the object.

When the own vehicle 40 is approaching a junction or an intersection and an object is recognized between the own lane 90 and the lane 91 (second lane), the object recognition device 2 recognizes this object as an obstacle 92 . (See FIG. 6). Then, as shown in FIG. 6, the object recognition device 2 draws a blind spot boundary on a straight extension line 93 connecting the detection point 80 at the end of the clustered obstacle 92 and the own vehicle 40 (sensor 10). directly. As described above, according to the object recognition device 2 according to the second embodiment, the blind spot boundary is set directly only when the host vehicle 40 is approaching a junction (or intersection) with poor visibility.

Further, as shown in FIG. 7, the object recognition device 2 extracts a detection point 89 farthest from the blind spot boundary on the lane 91 and tracks the extracted detection point 89 . Thus, the object recognition device 2 can recognize an object (another vehicle 62) related to the running of the own vehicle 40 by tracking the detection point 89 on the lane 91 that merges with the own lane 90. FIG.

The object recognition device 2 also decomposes the velocity vector of the detection point 89 into a velocity component in the traveling direction of the lane 91 and a velocity component in a direction perpendicular to the traveling direction. Thereby, the object recognition device 2 can accurately calculate the speed information (the speed component in the traveling direction of the lane 91) related to the running of the own vehicle 40. FIG.

Each function described in the above embodiments may be implemented by one or more processing circuits. Processing circuitry includes programmed processing devices, such as processing devices that include electrical circuitry. Processing circuitry also includes devices such as application specific integrated circuits (ASICs) and circuit components arranged to perform the described functions. Also, the object recognition device 1 and the object recognition device 2 can improve the function of the computer.

While embodiments of the present invention have been described above, the discussion and drawings forming part of this disclosure should not be construed as limiting the invention. Various alternative embodiments, implementations and operational techniques will become apparent to those skilled in the art from this disclosure.

1, 2 Object recognition device 10 Sensor 11 Map database 12 GPS receiver 20 Controller 21 Blind spot boundary calculator 22 Object recognition unit 23 Driving support unit 30 Display

Claims (4)

A sensor mounted on the own vehicle is used to detect a plurality of detection points on the surface of an object around the own vehicle, the plurality of detection points detected by the sensor are clustered, and a clustered cluster is obtained. is an object recognition method for recognizing as the object,
Calculate the position of the blind spot boundary indicating the boundary between the detectable range of the sensor and the undetectable range of the sensor,
get the map database,
obtaining the position of the own vehicle;
determining whether the own vehicle is approaching a junction based on the map database and the position of the own vehicle;
when the own vehicle is approaching the merging point, setting the blind spot boundary on a second lane that merges with the first lane on which the own vehicle travels;
determining that the object is an obstacle when the own vehicle is approaching the merging point and the object is recognized between the first lane and the second lane;
setting the blind spot boundary on an extension of a straight line connecting the clustered detection point of the edge of the obstacle and the sensor;
determining whether the cluster touches the blind spot boundary;
Extracting a detection point farthest from the blind spot boundary from among a plurality of detection points included in a cluster that is in contact with the blind spot boundary,
An object recognition method characterized by tracing extracted detection points. 2. The object recognition method according to claim 1 , wherein a detection point farthest from the blind spot boundary is extracted on the second lane, and the extracted detection point is tracked. 3. The object recognition according to claim 2 , wherein the velocity vector of the detection point farthest from the blind spot boundary is decomposed into a velocity component in the traveling direction of the second lane and a velocity component in a direction perpendicular to the traveling direction. Method. A sensor mounted on the own vehicle is used to detect a plurality of detection points on the surface of an object around the own vehicle, the plurality of detection points detected by the sensor are clustered, and a clustered cluster is obtained. is an object recognition device comprising a controller that recognizes as the object,
The controller is
Calculate the position of the blind spot boundary indicating the boundary between the detectable range of the sensor and the undetectable range of the sensor,
get the map database,
obtaining the position of the own vehicle;
determining whether the own vehicle is approaching a junction based on the map database and the position of the own vehicle;
when the own vehicle is approaching the merging point, setting the blind spot boundary on a second lane that merges with the first lane on which the own vehicle travels;
determining that the object is an obstacle when the own vehicle is approaching the merging point and the object is recognized between the first lane and the second lane;
setting the blind spot boundary on an extension of a straight line connecting the clustered detection point of the edge of the obstacle and the sensor;
determining whether the cluster touches the blind spot boundary;
Extracting a detection point farthest from the blind spot boundary from among a plurality of detection points included in a cluster that is in contact with the blind spot boundary,
An object recognition device characterized by tracking extracted detection points.

Images