JP7240886B2 - SELF-LOCATION ERROR ESTIMATION METHOD AND SELF-LOCATION ERROR ESTIMATION DEVICE - Google Patents

Description

The present invention relates to a self-position error estimation method and a self-position error estimation device.

Conventionally, a plurality of targets (also called landmarks) around the vehicle are imaged by an imaging means provided in the vehicle, and the positions of the plurality of targets on the imaged image and the positions of the targets stored in advance on the map are captured. A self-position detecting device is known that estimates its own position by matching the positions of a plurality of targets. (Patent document 1).

JP 2012-127896 A

However, as in the technique described in Patent Literature 1, self-position estimation errors inevitably occur in a self-position detecting device that uses target positions on an image. Therefore, in a self-position detection device that estimates the self-position based on the positions of a plurality of picked-up targets, accurate estimation of the self-position detection error has been desired.

An object of the present invention is to provide a self-position error estimation method and a self-position error estimation device capable of improving the accuracy of self-position detection error estimation.

A self-position error estimation method according to an aspect of the present invention sets a course in which a vehicle passes through a predetermined area on a road on which at least two targets can be detected, and the vehicle moves through the area along the course. A self-position detection error is estimated based on the number of pixels between the two targets on the camera image detected when they pass.

According to the present invention, the accuracy of estimating self-location detection errors can be improved.

FIG. 1 is a schematic configuration diagram of a self-position error estimation device according to an embodiment of the present invention. FIG. 2 is a diagram illustrating an embodiment of a self-location error estimation method. FIG. 3 is a diagram for explaining the number of pixels between targets. FIG. 4 is a flow chart explaining an operation example of the self-position error estimation device according to the embodiment of the present invention. FIG. 5 is a diagram for explaining an embodiment of a self-position error estimation method. FIG. 6 is a diagram for explaining the number of pixels between targets. FIG. 7 is a diagram for explaining an embodiment of a self-position error estimation 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.

(Configuration of self-position error estimation device)
The configuration of self-position error estimation device 100 will be described with reference to FIG. As shown in FIG. 1, the self-position error estimation device 100 includes a self-position detection unit 10, a map database 20, a camera 30, a controller 40, and a storage device 50.

The self-position error estimation device 100 may be installed in a vehicle having an automatic driving function, or may be installed in a vehicle without an automatic driving function. In addition, the self-position error estimation device 100 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 map database 20 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 20 also stores map information including the number of lanes of roads, road information including road boundaries, target information including position information of targets (landmarks), and the like. The map database 20 outputs map information to the controller 40 in response to requests from the controller 40 . Various data such as road information and target information are not necessarily acquired from the onboard map database 20, and may be acquired using vehicle-to-vehicle communication or road-to-vehicle communication. That is, when various data such as road information and target information are stored in an externally installed server, the controller 40 may acquire these data from the server at any time through communication. Also, the controller 40 may periodically acquire the latest map information from an externally installed server and update the map information it holds.

The camera 30 is mounted on the own vehicle and photographs the surroundings of the own vehicle. The camera 30 has an imaging device such as a CCD (charge-coupled device) or a CMOS (complementary metal oxide semiconductor). The camera 30 has an image processing function, and detects white lines, roadsides, targets, etc. from captured images (camera images). A target is, for example, an object provided on a road or sidewalk, such as a traffic light, a utility pole, or a traffic sign. Note that the target is not limited to these, and may be any object other than a moving object whose position is fixed, such as a building. The camera 30 outputs captured images, image analysis results, and the like to the controller 40 .

The self-position detection unit 10 detects the self-position, which is the position of the vehicle, from images captured by the camera 30 (hereinafter referred to as camera images or simply images) and target information stored in the map database 20 . Specifically, at least two targets (landmarks) are extracted from the camera image, and a specific target corresponding to the above-extracted targets is identified from a plurality of targets stored in the map database 20. do. Based on the position of the extracted target on the image and the position of the specific target included in the target information stored in the map database 20, the self-position on the map is detected. Note that the method of detecting the self-position in the self-position detecting section 10 is not limited to this. For example, the positions on the image of at least two targets extracted when the own vehicle traveled the same point in the past and the self-position detected using the extracted positions on the image of the target are stored in association with each other. Then, the next time the vehicle travels at the same point, the self-position may be detected based only on the position of the target on the image. That is, as long as the self-position detection unit 10 detects the self-position based on the positions of at least two target images captured by the camera 30, the method of detecting the self-position can be changed as appropriate. .

The controller 40 is a general-purpose microcomputer having a CPU (Central Processing Unit), memory, and an input/output unit. A computer program for functioning as the self-position error estimation device 100 is installed in the microcomputer. By executing the computer program, the microcomputer functions as a plurality of information processing circuits included in self-position error estimation device 100 . Here, an example of realizing a plurality of information processing circuits provided in the self-position error estimation device 100 by software will be shown. It is also possible to construct an information processing circuit. Also, a plurality of information processing circuits may be configured by individual hardware. The controller 40 includes a plurality of information processing circuits including an area setting unit 41, a course setting unit 42, a passing time calculating unit 43, a passing error estimating unit 44, a self-position error estimating unit 45, and a total error amount calculating unit 46. And prepare.

The area setting unit 41 uses the travel data of the own vehicle to set an area in which a larger number of pixels between targets on the image can be obtained than when the own vehicle travels in the center of the lane. In this embodiment, this area is set when offline. Off-line means a time when automatic driving is not being executed. The area set by the area setting unit 41 is stored in the storage device 50 . The number of areas set by the area setting unit 41 may be one or plural. A plurality of areas set by the area setting unit 41 may be set continuously along a course set by the course setting unit 42, which will be described later.

The course setting unit 42 reads the area from the storage device 50 when online. Next, the course setting unit 42 sets a course (route) along which the vehicle passes through the area. The online time refers to the time when automatic driving is being performed. The passage time calculator 43 calculates the time required for the vehicle to pass through the area (passage time). The passage error estimator 44 estimates a self-position detection error (hereinafter also referred to as a self-position error) when the vehicle passes through the area. The self-position error estimator 45 estimates the self-position error until the vehicle passes through the area. The total error calculator 46 calculates the total error of the self-position. The course set by the course setting unit 42 is a course passing through a predetermined area on the road where at least two targets can be detected.

Next, a method of estimating the self-position error during automatic operation will be described with reference to FIGS.

In autonomous driving, the vehicle is generally controlled to run in the center of the lane. As an example, as shown in Course 1 in FIG. 2, the own vehicle 60 is controlled to run in the center of the lane during automatic driving. Course 1 is a route passing through the center of the lane shown in FIG.

Accurate self-position detection is required in automatic driving. As an example of the self-position detection method, there is a method using targets 70 and 71 shown in FIG. Since the positions of the targets 70 and 71 on the map are fixed, the self-position is detected by using the distance and direction from the own vehicle 60 to the targets 70 and 71 . That is, the self position can be detected by matching the targets 70, 71 on the image with the targets 70, 71 on the map.

Specifically, for example, from the positions of the targets 70 and 71 on the image, the directions of the targets 70 and 71 with respect to the imaging direction of the camera 30 (directions of the targets 70 and 71 with respect to the own vehicle) are specified. That is, the angle formed by the line segment connecting the own vehicle and the target 70 and the line segment connecting the own vehicle and the target 71 is the distance (the number of pixels) between the target 70 and the target 71 on the image. handle. Therefore, the directions of the targets 70 and 71 with respect to the imaging direction of the camera 30 (the directions of the targets 70 and 71 with respect to the own vehicle) are specified from the distance between the targets 70 and 71 on the image. Therefore, if the targets 70 and 71 are specified on the map, the self-position is detected by known triangulation, for example, from the positions of the targets 70 and 71 on the map and the direction of the vehicle. The self-position detected using the positions of the targets 70 and 71 on such camera images contains an error. The self-position error may be represented by a covariance matrix obtained by applying the measurement error of the camera 30 to a normal distribution. Also, the self-location error may be represented by a likelihood distribution when matching the white line detection result and the map data.

The self-position error may change according to the passing position of the vehicle 60 on the lane. The passing position of the own vehicle 60 here includes the position where the own vehicle 60 (camera 30 ) detects the targets 70 and 71 .

There is a possibility that the self-position error will change between when the vehicle 60 runs along the course 1 shown in FIG. 2 and when the vehicle 60 runs along the course 2 shown in FIG. This point will be described with reference to FIG. An image 90 shown in FIG. 3 is an image of the targets 70 and 71 captured by the camera 30 at the position P1 on the course 1 shown in FIG. An image 91 shown in FIG. 3 is an image of targets 70 and 71 captured by camera 30 at position P2 on course 2 shown in FIG.

As can be seen from images 90 and 91 , image 91 has more pixels between target 70 and target 71 than image 90 . As described above, when detecting the self-position using the positions of the targets 70 and 71, an error is included according to the distance between the targets 70 and 71 on the image. The distance between the targets 70 and 71 referred to here is the distance between the targets 70 and 71 in the horizontal direction on the image. The distance between the target 70 and the target 71 in the horizontal direction on the image is calculated with higher accuracy as the number of pixels between the target 70 and the target 71 in the horizontal direction on the image increases. That is, when own vehicle 60 runs along course 1 shown in FIG. 2 and when own vehicle 60 runs along course 2 shown in FIG. In the case, the distance between the target 70 and the target 71 is calculated more accurately. As a result, the self-position error is smaller when the vehicle 60 travels along the course 2 than when the vehicle 60 travels along the course 1 .

The details of the method for estimating the self-position error during automatic operation will be described with reference to the flowchart shown in FIG. In step S101, the region setting unit 41 uses the travel data of the vehicle 60 to compare the targets 70 and 71 in the lateral direction on the image with respect to the case where the vehicle 60 travels in the center of the lane. A region (region 80 in FIG. 2) in which a large number of pixels can be obtained is set. In the area 80 shown in FIG. 2, a larger number of pixels are acquired between the target 70 and the target 71 in the horizontal direction on the image than when the vehicle 60 runs in the center of the lane. As described above, the area 80 is set and stored in the storage device 50 when offline. When setting the area 80, travel data of other vehicles may be used. An appropriate region 80 is set by using more travel data including travel data of other vehicles. In the area 80, the error in the positions of the targets 70 and 71 captured by the camera 30 is a predetermined value ( (first predetermined value) or less. Thereby, the area 80 can have a predetermined range, and the self-position error can be kept small. Note that the targets 70 and 71 shown in FIG. 2 are utility poles, but are not limited to this. The targets 70, 71 may be road signs. Also, the area 80 may be an area in which the areas of the images of the targets 70 and 71 captured by the camera 30 at the position P1 shown in FIG. 2 are equal to or larger than a predetermined value (second predetermined value). This allows the region 80 to have a predetermined range and increases the action options for keeping the self-position error small.

The process proceeds to step S103, and the course setting unit 42 reads the area 80 from the storage device 50 when online. Next, the course setting unit 42 sets a course 2 passing through the area 80 as shown in FIG. 2, for example. Note that the course 2 is a route set by, for example, a navigation device, and is an arbitrary route.

The process proceeds to step S<b>105 , and the passage time calculator 43 calculates the time (passage time) required for the own vehicle 60 to pass through the area 80 along the course 2 . As an example, the passage time calculator 43 calculates the passage time based on the speed of the own vehicle 60 . When the speed of the own vehicle 60 is fast, the passage time becomes short. On the other hand, when the speed of own vehicle 60 is slow, the passage time is long. The longer the transit time, the more images acquired during the transit. The more images acquired during a pass, the more accurate the analysis of the images, and thus the smaller the self-location error. Therefore, self-position error estimation device 100 may decelerate self-vehicle 60 when self-vehicle 60 passes region 80 along course 2 in order to reduce self-position error. As a result, the time required for the vehicle 60 to pass through the area 80 along the course 2 is increased, and more images are acquired.

The process proceeds to step S<b>107 , and the passing error estimator 44 estimates the self-position error when the own vehicle 60 travels along the course 2 in the region 80 . As an example of the estimation method, the passing error estimating unit 44 creates and stores in advance an error model that models the relationship between the image and the error of the self-position. By comparing the image with the obtained image, the error generated at each point is calculated, and the calculated error is integrated for the time during which the vehicle 60 travels in the area 80, so that the vehicle 60 can move along the course 2 to the area. Estimate self-position error when driving 80 . The error model is a model including the shapes, positional information, etc. of the targets 70 and 71 shown in FIG.

That is, when the self-position is detected based on the positions of the targets 70 and 71 on the image as described above, the distance (the number of pixels) between the target 70 and the target 71 on the image determines the detection accuracy of the self-position. affects For this reason, a model formula including the distance on the image between the two targets (targets 70 and 71) and the error of the self-position is stored in advance, and the object at each point on the course 2 within the area 80 is stored. From the distance between target 70 and target 71 and the stored model formula, the error of the self-position at each point is estimated and integrated. Estimate the self-position error. As described above, the slower the time to pass through the region 80, that is, the slower the speed when traveling through the region 80, the smaller the self-position error. A model may be created in which the slower (or the longer the passing time) the smaller the error. The error of the self-position when the host vehicle 60 passes through the region 80 along the course 2 may hereinafter be expressed as a first error. Note that the passing error estimating unit 44 may model the measurement error of the camera 30 with a covariance matrix when fitted to a normal distribution, and then add the eigenvalues of the covariance matrix to estimate the first error.

The process proceeds to step S<b>109 , and the self-position error estimator 45 estimates the self-position error before the vehicle 60 travels through the area 80 . The self-vehicle 60 estimates the error of its own position even before traveling in the area 80, and the error is accumulated. Therefore, by accumulating and storing errors in the estimated self-position before the vehicle 60 runs through the area 80, the self-position generated before the vehicle 60 runs through the area 80 can be read out. can be estimated. The self-position error before the vehicle 60 travels through the area 80 may hereinafter be referred to as a second error. The self-position error estimator 45 estimates such a second error.

The process proceeds to step S111, and the error total amount calculator 46 calculates the sum of the first error and the second error. In other words, the total error calculator 46 calculates the total error of the self-position. When the self-position error is represented by a covariance matrix obtained by applying the measurement error of the camera 30 to a normal distribution, the total self-position error is the reciprocal of the sum of the eigenvalues.

At position P3 on course 3 shown in FIG. 5, image 92 has more pixels between targets 70 and 71 than images 90 and 91, as shown in FIG. That is, when own vehicle 60 runs along course 1 shown in FIG. 5, when own vehicle 60 runs along course 2 shown in FIG. The distance between the target 70 and the target 71 is calculated most accurately when the own vehicle 60 runs along the course 3, compared to the case where the vehicle 60 runs along the course. In other words, the self-position error is smaller when the vehicle 60 travels along the course 3 than when the vehicle 60 travels along the courses 1 and 2 .

In this way, by estimating the self-position error when traveling in the region 80 for each of a plurality of courses in which the region 80 is traveled, for example, when traveling in the region 80 by automatic driving, the self-position error is the smallest. It can be used meaningfully for autonomous driving, such as selecting a course and driving.

As described above, according to the self-position error estimation device 100 according to the present embodiment, the following effects are obtained.

The self-position detection unit 10 extracts at least two targets (targets 70 and 71) from the image captured by the camera 30, and detects the self-position based on the positions of the extracted targets on the image. When self-position error estimation device 100 travels along course in area 80, self-position error estimating device 100 calculates the distance between two targets on the image captured at each point on the course. By estimating the self-position error at the points, a first error, which is the self-position error as the vehicle 60 travels along the course in the region 80, is estimated.

The distance between the target 70 and the target 71 on the image is calculated with higher accuracy as the number of pixels between the target 70 and the target 71 on the image increases. That is, when own vehicle 60 runs along course 1 shown in FIG. 2 and when own vehicle 60 runs along course 2 shown in FIG. In the case, the distance between the target 70 and the target 71 is calculated more accurately.

Therefore, according to the self-position error estimation device 100, when the own vehicle 60 travels along the predetermined course in the area 80, two images of the target on the image taken at each point on the course By estimating the self-position detection error based on the above distance, the self-position error can be estimated with high accuracy.

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 self-position error estimation device 100 can improve the functionality of a 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.

In the example shown in FIG. 2, two targets are observed from the area 80, but the number is not limited to this. Three or more targets may be observed from the area 80 . As the number of targets increases, the estimation accuracy of the self-position error can be improved.

Note that the self-position detection unit 10 (self-position detection device) may be configured by a general-purpose microcomputer having a CPU, memory, and input/output unit, like the controller 40 , or may be incorporated in the controller 40 .

10 self-position detector 20 map database 30 camera 40 controller 41 area setting unit 42 course setting unit 43 passage time calculator 44 passage error estimator 45 self-position error estimator 46 total error calculator 50 storage devices 70 and 71 target 80 Regions 90, 91, 92 Image 100 Self-position error estimator

Claims (4)

Self-position detection error in a self-position detection method that extracts at least two targets from an image captured by a camera mounted on the vehicle and detects the self-position based on the positions of the extracted targets on the image. An estimation method comprising:
setting a course in which the own vehicle passes through a predetermined area on a road on which at least two targets can be detected;
and estimating a detection error of the self-position based on the number of pixels between the two targets on the camera image detected when the vehicle passes through the area along the course. self-position error estimation method. 2. The self-position error estimation method according to claim 1, wherein said area is an area in which three or more targets can be detected. 2. A self-position error estimation method according to claim 1, wherein said area is an area of said two targets on said camera image having an area equal to or larger than a second predetermined value. Detecting the self-position in a self-position detection device that extracts at least two targets from an image taken by a camera mounted on the vehicle and detects the self-position based on the positions of the extracted targets on the image. A self-position error estimation device for estimating an error,
a controller for setting a predetermined area on the road where at least two targets can be detected;
The controller is
setting a course for the own vehicle to pass through the area;
and estimating a detection error of the self-position based on the number of pixels between the two targets on the camera image detected when the vehicle passes through the area along the course. self-position error estimator.

Images