JP2017181476A - Vehicle location detection device, vehicle location detection method and vehicle location detection-purpose computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a vehicle location detection device that can accurately estimate a location of an own vehicle at a low processing load.SOLUTION: A vehicle location detection device includes: a particle generation unit (23) that generates a plurality of particles expressing a luminance change in a direction crossing a line according to a category of the line as to each line to be marked on a road extracted from map information in accordance with a prediction location at a current time of an own vehicle (10); a projection unit (24) that projects each particle onto an image of the current time obtained by an imaging unit (2) loaded in the vehicle (10) as changing the prediction location within an expected error range; a selection unit (25) that selects the particle in which a degree of matching to the image satisfies a prescribed matching condition of each particle; and a location update unit (26) that obtains an estimation location of the vehicle (10) in accordance with a total sum of errors when projecting each of the selected particles onto the line to be marked on the corresponding road on the basis of the estimation location of the vehicle (10) at the current time.SELECTED DRAWING: Figure 4

Description

  The present invention relates to a vehicle position detection device, a vehicle position detection method, and a vehicle position detection computer program that detect the position of a vehicle using an image and a map taken by a camera mounted on the vehicle.

  2. Description of the Related Art Conventionally, for navigation or driving assistance, a position of the host vehicle while traveling is detected using a Global Positioning System (GPS) signal. However, when there is a structure that blocks the positioning signal from the GPS positioning satellite, such as a high-rise building, near the vehicle, it becomes difficult to receive the positioning signal in the vehicle, and the position of the host vehicle cannot be detected accurately. There was a thing. Therefore, a method for improving the accuracy of detecting the position of the host vehicle by associating an image of the surroundings of the vehicle, which is captured by the in-vehicle camera, with information such as a map prepared in advance has been proposed (for example, a patent). References 1-3 and Non-Patent References 1 and 2).

  For example, the position positioning device described in Patent Document 1 specifies the current position based on a road marking in an image captured by an imaging unit that captures the front of the vehicle.

  In addition, the moving body position measuring device described in Patent Document 2 includes feature information extracted from an image captured by an imaging unit mounted on a moving body in the vicinity of a reference point on the map, and the vicinity of the reference point on the map. The current position of the moving object is estimated in accordance with the positions of a plurality of features near the reference point estimated by matching with the feature information extracted from the image.

  Moreover, the method described in Patent Document 3 determines the position of the vehicle with respect to the optical index identified in the image by applying template matching to the image data in the traveling direction of the vehicle.

  Furthermore, the method described in Non-Patent Document 1 detects the position of the host vehicle by matching a map including line segments representing curbs and road markings and feature points extracted from an image obtained by a stereo camera.

  Furthermore, the method described in Non-Patent Document 2 generates a high-resolution environment map using information acquired by a vehicle-mounted device, such as GPS, IMU, odometry, and LIDAR. This method uses a particle filter that correlates the LIDER measurement value with the map when positioning the vehicle on these maps.

JP 2007-108043 A JP 2012-127896 A JP 2005-136946 A

Schreiber et al., "LaneLoc: Lane Marking based Localization using Highly Accurate Maps", Intelligent Vehicles Symposium (IV), 2013 IEEE, IEEE, 2013 Lenvision et al., “Map-Based Precision Vehicle Localization in Urban Environments”, Robotics: Science and Systems (RSS), 2007

  However, the device described in Patent Document 1 may not be able to accurately identify the position of the host vehicle when it is difficult to detect feature points of road markings due to environmental influences during shooting. In particular, the device described in Patent Document 1 extracts a road marking end point or the like as a feature point, and uses the feature point to determine the current position of the vehicle. Even if it is slightly drowned, the position of the extracted feature point is shifted, and the detection accuracy of the position of the host vehicle is lowered.

  The device described in Patent Document 2 uses an image photographed in the vicinity of a reference point on a map for specifying the position of the host vehicle, but rebuilding a building near the reference point or an environment at the time of photographing (for example, , Weather, shooting time, etc.), even if the image was taken at the same location, it appears in the scene taken in the image taken at the time of pre-registration and in the image taken when the vehicle position was detected The scenery can vary greatly. In such a case, there is a high possibility that matching between images will fail, and therefore the position of the host vehicle may not be accurately identified.

  Furthermore, the method described in Patent Document 3 uses template matching. In template matching, since it is necessary to repeatedly perform matching calculation while changing the position between the image to be compared and the template, a large amount of calculation is required for detecting the position of the host vehicle. Further, in the method described in Non-Patent Document 1, different algorithms are used for curb detection and road marking detection, and a stereo image is used for these detections. Therefore, in these methods, hardware resources necessary for detecting the position of the own vehicle increase, or time is required for detecting the position of the own vehicle, so that the update interval of the position of the own vehicle becomes long. . Furthermore, in the method described in Non-Patent Document 2, a laser ranging system is required for detecting the position of the host vehicle, and the cost of the entire system increases.

  Accordingly, an object of the present invention is to provide a vehicle position detection device, a vehicle position detection method, and a vehicle position detection computer program that can accurately estimate the position of the host vehicle with a low processing load.

According to the first aspect of the present invention, a vehicle position detecting device is provided as one aspect of the present invention. The vehicle position detection device includes a storage unit (41) that stores map information indicating the position of a line marked on a road and the type of the line, and the current vehicle position estimated from the position immediately before the vehicle (10). Lines marked on the road are extracted from the map information according to the predicted position at the time, and for each of the extracted lines, particles representing the luminance change in the direction crossing the line are determined according to the type of the line. A particle generation unit (23) that generates a plurality of particles while changing the position according to the direction along the line, and a predicted position of each of the plurality of particles for each of the extracted lines is changed within an assumed error range. A projection unit (24) that obtains a position with reference to the vehicle (10) and projects the current position image obtained by the imaging unit (2) mounted on the vehicle (10) according to the position. A selection unit (25) that selects a particle satisfying a predetermined matching condition among a plurality of particles for each of the extracted lines, and a vehicle ( A position updating unit (26) for obtaining an estimated position of the vehicle (10) according to a sum of errors when projected onto a line marked on the corresponding road based on the estimated position of (10).
Since the vehicle position detection device according to the present invention has the above configuration, the position of the host vehicle can be accurately estimated with a low processing load.

According to the second aspect of the present invention, the position updating unit (26) is configured to minimize the sum of errors when each of the selected particles is projected onto a line marked on the corresponding road. It is preferable to obtain the estimated position of 10).
By having this configuration, the vehicle position detection device can estimate the position of the host vehicle more accurately.

According to a third aspect of the present invention, the position updating unit (26) calculates a sum of errors when each of the selected particles is projected onto a line marked on the corresponding road, and a plurality of past times. The current time so that the sum of the error between the vehicle position at the next time predicted from the estimated position of the vehicle (10) at each time using the odometry information and the estimated position of the vehicle at the next time is minimized. It is preferable to obtain the estimated position of the vehicle (10) at each of a plurality of past times.
By having this configuration, the vehicle position detection apparatus also uses odometry information, so that the position of the host vehicle can be estimated more accurately.

Furthermore, according to the description of claim 4, the particle generation unit (23) indicates on the road corresponding to the case where there is a dotted line or broken line block when the extracted line type includes a line including a dotted line or a broken line. A first particle representing a luminance change in a direction crossing the line to be generated and a second particle representing a luminance change in a direction crossing the line marked on the road corresponding to the case where there is no dotted or broken line block It is preferable to do.
With this configuration, the vehicle position detection device can generate particles that are highly likely to coincide with a line that is actually marked on the road.

Further, according to the fifth aspect of the present invention, the imaging unit (2) is a stereo camera including two cameras, and the selection unit (25) is the first one obtained by one of the two cameras at the current time. It is preferable to select a particle that satisfies the matching condition on one image and satisfies the matching condition on the second image obtained at the current time by the other of the two cameras.
By having this configuration, the vehicle position detection device can reduce the possibility of erroneously selecting particles projected on a place where the road surface is hidden due to occlusion by another vehicle or the like.

According to the description of claim 6, a vehicle position detection method is provided as another embodiment of the present invention. In this vehicle position detection method, a line marked on a road is extracted from map information according to a predicted position at the current time of the vehicle estimated from a position immediately before the vehicle (10), and each extracted line is detected. In accordance with the type of the line represented in the map information, a step of generating a plurality of particles representing luminance changes in the direction crossing the line while changing the position according to the direction along the line, and the extracted line For each of a plurality of particles, a position based on the vehicle (10) is obtained while changing the predicted position within an assumed error range, and an imaging unit mounted on the vehicle (10) according to the position Of the plurality of particles for each of the extracted lines and the step of projecting on the current time image obtained in (2), the degree of coincidence with the image is a predetermined one. According to the step of selecting particles that satisfy the condition and the sum of errors when each of the selected particles is projected onto a line marked on the corresponding road based on the estimated position of the vehicle (10) at the current time And obtaining an estimated position of the vehicle (10).
Since the vehicle position detection method according to the present invention has the above-described configuration, the position of the host vehicle can be accurately estimated with a low processing load.

According to the sixth aspect of the present invention, a computer program for detecting a vehicle position is provided as another aspect of the present invention. This computer program for vehicle position detection extracts a line that is marked on the road according to the predicted position at the current time of the vehicle estimated from the position immediately before the vehicle (10) from the map information. For each, a step of generating a plurality of particles representing luminance changes in the direction crossing the line according to the type of the line represented in the map information while changing the position according to the direction along the line, and extraction For each of the plurality of particles for each of the lines, a position based on the vehicle (10) is obtained while changing the predicted position within an assumed error range, and the position is mounted on the vehicle (10) according to the position. The step of projecting on the image at the current time obtained by the imaging unit (2), and the image among the plurality of particles for each of the extracted lines A step of selecting particles satisfying a predetermined degree of coincidence, and projecting each of the selected particles onto a line marked on the corresponding road based on the estimated position of the vehicle (10) at the current time And a step of obtaining an estimated position of the vehicle (10) in accordance with the sum of the errors of the processor (43) included in the vehicle (10).
The computer program for detecting a vehicle position according to the present invention has the above configuration, so that the position of the host vehicle can be accurately estimated with a low processing load.

  The reference numerals in parentheses attached to the above-described parts are examples that show the correspondence with specific means described in the embodiments described later.

1 is a schematic configuration diagram of a vehicle position detection system according to one embodiment of the present invention. It is a figure which shows the relationship between a map coordinate system, a vehicle coordinate system, and a camera coordinate system. (A)-(d) is a figure which shows an example of map information. It is a functional block diagram of the control part of a vehicle position detection system. It is a figure which shows an example of a particle in case the kind of map line segment is "white solid line (thin)." It is a figure which shows an example of a particle in case the kind of map line segment is "white solid line (thin) + right-side deceleration indication". It is a figure which shows an example of a particle in case the kind of map line segment is "white dotted line (thin) + both-sides deceleration indication". It is a figure which shows an example of the image which each particle projected. It is a figure showing an example of the error ellipse equivalent to the covariance matrix about each selected particle in a vehicle coordinate system. It is an operation | movement flowchart of a vehicle position detection process. It is a figure which shows an example of the relationship between the particle | grains produced | generated by a modification, and the line marked on a road.

Hereinafter, the vehicle position detection system will be described with reference to the drawings.
This vehicle position detection system refers to map information including information on types of lines marked on roads such as lane markings or road markings, and predicted positions at the current time estimated from the position of the vehicle itself immediately before A plurality of particles having a luminance profile representing a luminance change in a direction crossing the line according to the type of the line are generated along a line on the map around the area. The vehicle position detection system converts each particle represented in the coordinate system on the map into a position in a coordinate system based on the vehicle according to the predicted position and predicted direction at the current time of the host vehicle. Projecting onto an image obtained by a camera mounted on the vehicle. The vehicle position detection system selects a particle whose degree of coincidence with the image satisfies a predetermined coincidence among the particles projected on the image, and the projection error of the selected particle onto the line on the map is small. In this way, the position and direction of the host vehicle at the current time are estimated.

  FIG. 1 is a schematic configuration diagram of a vehicle position detection system according to one embodiment. As shown in FIG. 1, the vehicle position detection system 1 is mounted on a vehicle 10 and includes a camera 2 and a controller 4. The camera 2 and the controller 4 are connected to each other by a control area network (hereinafter referred to as CAN) 3. In FIG. 1, for convenience of explanation, each component of the vehicle position detection system 1 and the shape, size, and arrangement of the vehicle 10 are different from actual ones.

  The camera 2 is an example of an imaging unit, captures a front area of the vehicle, and generates an image of the front area. For this purpose, the camera 2 includes a two-dimensional detector composed of an array of photoelectric conversion elements having sensitivity to visible light, such as CCD or C-MOS, and the ground existing in front of the vehicle 10 on the two-dimensional detector. Alternatively, an imaging optical system that forms an image of a structure or the like is included. The camera 2 is disposed in the vehicle interior of the vehicle 10 such that the optical axis of the imaging optical system is substantially parallel to the ground and faces the front of the vehicle 10, for example. The camera 2 captures images at regular time intervals (for example, 1/30 second), and generates a color image capturing the front area of the vehicle 10. Note that the camera 2 may include a two-dimensional detector having sensitivity to near infrared light, and generate a monochrome image corresponding to the illuminance of the near infrared light within the imaging range.

FIG. 2 is a diagram illustrating the relationship among the map coordinate system, the vehicle coordinate system, and the camera coordinate system. In this embodiment, for convenience, a map coordinate system (Xm, Ym, Zm) having an arbitrary position on the map as an origin, a vehicle coordinate system (Xv, Yv, Zv) having the vehicle 10 as an origin, and the camera 2 are provided. Use the camera coordinate system (Xc, Yc, Zc) as the origin. In this embodiment, the vehicle coordinate system (Xv, Yv, Zv) has a midpoint between the left and right rear wheels of the vehicle 10 and a point on the ground as the origin. The traveling direction of the vehicle is the Zv axis, the direction orthogonal to the Zv axis and parallel to the ground is the Xv axis, and the vertical direction is the Yv axis. In the map coordinate system (Xm, Ym, Zm), the Xm axis and the Zm axis are set in a plane parallel to the ground without inclination, and the Ym axis is set in the vertical direction with respect to the ground. In the camera coordinate system (Xc, Yc, Zc), for the sake of simplicity of explanation, the center of the imaging surface is located above the origin of the vehicle coordinate system along the vertical direction and at a height at which the camera 2 is installed. Assuming that there is, the center of the imaging surface is the origin. Similarly to the vehicle coordinate system, the traveling direction of the vehicle 10 is the Zc axis, the direction orthogonal to the Zc axis and parallel to the ground is the Xc axis, and the vertical direction is the Yc axis.
In addition, although the position where the camera 2 is actually attached may be shifted from above the midpoint between the left and right rear wheels of the vehicle 10, this shift may be corrected by a simple parallel movement.

  The vehicle position detection system 1 may include a rear camera that captures the rear region of the vehicle as an image capturing unit instead of or together with the camera that captures the front region of the vehicle 10.

  The camera 2 sequentially transmits the generated image to the controller 4. In addition, when the camera which image | photographs the front area | region of the vehicle 10 and the camera which image | photographs the back area | region of the vehicle 10 are attached, the controller 4 selects only the image from the camera which image | photographs the direction which the vehicle 10 is advancing. May be acquired automatically. For this purpose, the controller 4 acquires a shift position signal representing the position of the shift lever from the electronic control unit (ECU) 11 of the vehicle 10 via the CAN 3. And the controller 4 acquires an image from the camera which image | photographs the front area | region of the vehicle 10, when the shift position signal becomes the drive position etc. which show that the vehicle 10 moves forward. On the other hand, when the shift position signal is a reverse position indicating that the vehicle 10 moves backward, the controller 4 acquires an image from a camera that captures the rear region of the vehicle 10.

  The controller 4 is an example of a vehicle position detection device, and includes a storage unit 41, a communication unit 42, and a control unit 43. The storage unit 41 includes, for example, a semiconductor memory such as an electrically rewritable nonvolatile memory and a volatile memory. The storage unit 41 stores various programs for controlling the vehicle position detection system 1, map information, camera position information such as the height of the camera 2 from the ground and the optical axis direction, the focal length and the angle of view of the imaging optical system. And various parameters such as camera parameters and temporary calculation results by the control unit 43 are stored. Further, the storage unit 41 is a parameter for correcting distortion aberration caused by the imaging optical system of the camera 2, for example, a correction amount of distortion aberration for each pixel (that is, an amount of movement of the pixel on the image for canceling the distortion aberration). And the moving direction) may be stored.

Hereinafter, map information used for vehicle position detection will be described.
Fig.3 (a)-FIG.3 (d) are figures which show an example of map information. The map information 300 shown in FIG. 3A is associated with latitude / longitude information indicating the position of the area represented by the map information. The map information 300 includes information related to the line 301 marked on the road such as a lane marking represented by a white line, a yellow line, or the like.

There are several types of lines that are marked on an actual road. For example, a line 311 indicated by a dotted line in FIG. 3B is represented by a combination of a solid white line and a broken-line deceleration sign provided on one side. A line 312 indicated by a dotted line in FIG. 3C is represented by a combination of a dotted white line and deceleration signs provided on both sides thereof. In addition, a line 313 indicated by a dotted line in FIG. 3D is represented only by a relatively thick solid white line. Thus, there are various types of lines that are marked on the road. However, in the map information 300, for example, a line on each road is represented by one solid line, and a label indicating the type of the line may be attached to each line. For example, in the example shown in FIG. 3B, the line 311 is represented by a solid line 321 and a label of (white line (thin) + right deceleration marking). In the example shown in FIG. 3C, the line 312 is represented by a solid line 322 and a label of (white dotted line (thin) + both side deceleration marking). In the example shown in FIG. 3D, the line 313 is represented by a solid line 323 and a label of a white line (thick). Further, the map information 300 includes the three-dimensional coordinates of both end points of each line marked on the road.
In the following, the line marked on the road included in the map information is referred to as a map line segment.

  The communication unit 42 includes a communication interface that communicates with various sensors such as the camera 2, the ECU 11, and a wheel speed sensor (not shown) through the CAN 3, and a control circuit thereof. The communication unit 42 receives an image from the camera 2 and passes the image to the control unit 43. In addition, the communication unit 42 obtains the speed and movement amount of the vehicle 10 from the ECU 11 as odometry information or obtains the wheel speed from the wheel speed sensor for each period of executing the position estimation process. Pass to part 43.

The control unit 43 includes one or a plurality of processors (not shown) and their peripheral circuits. And the control part 43 controls the vehicle position detection system 1 whole.
FIG. 4 shows a functional block diagram of the control unit 43. As illustrated in FIG. 4, the control unit 43 includes an initial value setting unit 21, a predicted position calculation unit 22, a particle generation unit 23, a projection unit 24, a selection unit 25, and a position update unit 26. Each of these units included in the control unit 43 is implemented as a functional module realized by a computer program executed on a processor included in the control unit 43, for example.

The initial value setting unit 21 sets the initial value of the position of the host vehicle when starting the detection of the position of the vehicle 10 according to the present embodiment, that is, the detection of the position of the host vehicle. For example, the initial value setting unit 21 obtains a position obtained from the positioning signal when the GPS positioning signal was last received from, for example, a navigation system, and based on the position, the initial value of the position of the host vehicle is obtained. Set the value. Then, the obtained position and direction are set to the initial value of the position of the host vehicle.
Note that the initial value set by the initial value setting unit 21 may be a value obtained with detection accuracy lower than the detection accuracy of the position of the host vehicle according to the present embodiment.

In this embodiment, the initial value μ _t = (x _t , z _t , θ _t ) ^t of the position of the host vehicle at time t is expressed by the following equation according to a Gaussian distribution.
Here, (x _t , z _t ) is the position of the vehicle 10 in the x direction and the z direction in the map coordinate system obtained from the positioning signal or the like when the time for calculating the initial value is t, that is, Indicates the position of the host vehicle. theta _t is at time t, represents the orientation of the vehicle 10. In addition, the control part 43 receives the signal from the sensor (not shown) which measures the azimuth | direction of the traveling direction of the vehicle 10 which the vehicle 10 has about the direction of the vehicle 10 via CAN3 and the communication part 42, for example. Is obtained. Alternatively, the control unit 43 calculates the direction of the host vehicle calculated from the change in the position of the host vehicle obtained from the GPS positioning signals received at different times immediately before starting the detection of the position of the host vehicle. You may receive from. Further, in (1), the sigma _t, the covariance matrix. Further, the position in the vertical direction, that is, the altitude y _t is not included in the equation (1), but can be calculated by the following equation.
Here, a _x ^t , a _z ^t , and a _c ^t are constants, respectively.For example, assuming that the map coordinates (x _t , z _t ) are planar, the map coordinates (x _t , z _When the map coordinates (x _i , y _i , z _i ) of each equidistant point on each map segment included in a predetermined range (for example, within 10m) around _t ) are substituted into equation (2) For example, the least square method is used so that the square error is minimized.

  The initial value of the position of the host vehicle set by the initial value setting unit 21 is stored in the storage unit 41.

The predicted position calculation unit 22 predicts the host vehicle at the current time based on the initial value of the position of the host vehicle or the position of the host vehicle immediately before obtained by the position update unit 26 and the wheel speed of the vehicle 10. Calculate the position.
In the present embodiment, the predicted position of the host vehicle at the current time t is expressed as a Gaussian distribution with an average value μ _p ^t and a covariance Σ _p ^t . Therefore, the predicted position calculation unit 22 performs covariance with the average value μ _p ^t = (x _p ^t , z _p ^t , θ _p ^t ) of the predicted position and predicted direction in the horizontal direction of the host vehicle at the current time t according to the following equation. The matrix Σ _p ^t is calculated.
Here, the Gaussian distribution represented by the average value μ _t-1 and the covariance Σ _t-1 represents the estimated position of the host vehicle at time (t-1). When the current time t is the first time when the control unit 43 detects the position of the host vehicle, N (x _t−1 , z _t−1 , θ _t−1 | μ _t−1 , Σ _t−1 ) is The initial value of the position of the host vehicle set by the initial value setting unit 21. Further, when the control unit 43 detects the position of the host vehicle at time (t−1), N (x _t−1 , z _t−1 , θ _t−1 | μ _t−1 , Σ _t−1 ) Is the estimated position of the host vehicle detected by the control unit 43 at time (t-1).

In the equation (3), u _t = (v _t , ω _t ) ^t represents a control input determined from the wheel speed. v _t is the speed of the vehicle 10 at the current time t, the latest wheel speed v ^R _t of the right rear wheel and the left rear received from the wheel speed sensors attached to the left and right rear wheels via the communication unit 42. The average value (v ^R _t + v ^L _t ) / 2 of the wheel speed v ^L _t of the wheel is calculated. Further, ω _t is an angular velocity of the vehicle 10 at the current time t, and is ωt = (v ^R _t −v ^L _t ) / 2da. da is the distance between the left and right rear wheels.
Δt is a time interval between the current time t and the immediately preceding time (t−1), that is, a time interval at which the control unit 43 performs the detection processing of the host vehicle position. The matrix Q _t represents the degree of error included in the prediction of the position of the host vehicle based on the rear wheel speed. The elements σ _x , σ _z , σ _θ of the matrix Q _t are the standard deviation of the Xm-direction position error, the Zm-direction position error, and the direction error standard deviation in the map coordinate system, respectively. For example, σ _x = σ _z = 1 (m) and σ _θ = 10 (°) are set.
The predicted position calculation unit 22 may calculate the altitude y _t of the host vehicle at the current time t, for example, according to the equation (2).

The predicted position calculation unit 22 stores μ _p ^t and Σ _p ^t in the storage unit 41.

  The particle generation unit 23 is a particle that is a luminance pattern having a luminance profile that represents a luminance change in a direction across the map line segment included in the shooting range of the camera 2 at the predicted position based on the predicted position of the vehicle 10 at the current time. Is generated according to the type of the map line segment included in the map information.

  In the present embodiment, the particle generation unit 23 sets the generation range and density of the particles appropriately in a predetermined range around the predicted position (for example, within 30 m) and the camera 2 with the predicted position as a reference. Each map line segment within the imaging range is extracted from the map information, and each extracted map line segment is projected on an image obtained by the camera 2. In addition, for the map line segment where one end is within the predetermined range and the photographing range and the other end is out of the predetermined range or the photographing range, the particle generating unit 23 determines the boundary between the predetermined range and the photographing range. The point intersecting the nearer side from the predicted position is used as the other end of the map line segment, and the map line segment is projected on the image. Hereinafter, since the same processing is performed for each map line segment, the r-th map line segment will be described as an example.

The relationship between the point p _m = (x _m , y _m , z _m ) on the map expressed in the map coordinate system and the point u _c = (u, v) on the image is expressed by the following equation.
Here, p _v is a point of the vehicle coordinate system corresponding to the point p _m, p _c is a point in the camera coordinate system corresponding to the point p _m. The matrix {R _slope ^t R _vm ^t } represents a rotation component in the coordinate transformation from the vehicle coordinate system to the map coordinate system at time t. Here, the rotation matrix R _slope ^t representing the rotation component of the roll pitch is calculated from (a _x ^t , a _z ^t , a _c ^t ) in equation (2). Further, the translation sequence t _vm ^t = (x _vm ^t , y _vm ^t , z _vm ^t ) represents a translation component in the coordinate conversion from the vehicle coordinate system to the map coordinate system at time t. H (p _c , k _c ) represents a relationship with a point u _c on the image obtained by the camera 2 from the point p _c = (x _c , y _c , z _c ) in the camera coordinate system, and k _c is Represents a conversion parameter from the camera coordinate system to the image. When the distortion aberration of the camera 2 can be ignored, h (p _c , k _c ) is expressed by the following equation.

Particle generator 23, based on ^{_{μp t (x vm t, z}} vm t, θ vm t) Set, according to (4), both end points of the r th map line (m _s ^r, m _e ^r ) is projected onto the image obtained by the camera 2. Then, the particle generation unit 23 obtains the length l _img ^r of the map line segment on the image. The particle generator 23 divides each of the positions obtained by equally dividing the length l _img ^r of the map line segment by the total number of particles generated per map line segment (for example, 500 to 1000). A candidate for the position to be generated The particle generator 23 randomly selects a predetermined number of candidates from each candidate according to a uniform distribution. The particle generator 23 generates particles at a three-dimensional position in the map coordinate system corresponding to the selected candidate. Alternatively, the particle generator 23 may generate particles for all candidates.

  The particle generator 23 determines the particle type according to the type of the map line segment that generates the particle. Note that the types of map line segments are included in the map information as described above. At that time, depending on the type of the map line segment, the particle generation unit 23 generates a plurality of types of particles according to the probability corresponding to the type of the particle.

  FIG. 5 is a diagram illustrating an example of particles when the type of map line segment is “white line (thin)”. In FIG. 5, the horizontal axis represents the position along the direction perpendicular to the map line segment with the position of the solid white line as the origin, and the vertical axis represents the luminance. A profile 500 represents a luminance profile of particles to be generated. When the type of map line segment is “white line (thin)”, a single white line is marked on the road, so it crosses the map line segment regardless of the position in the direction along the map line segment. The luminance profile in the direction does not change. Therefore, in this case, the particle generation unit 23 has a high luminance only in the central portion corresponding to the position of the white line represented by the profile 500 and having the same width as the width of the white line with the occurrence probability p1 = 1.0. And generate particles with low brightness on both sides.

  FIG. 6 is a diagram illustrating an example of a particle when the type of the map line segment is “white line (thin) + right deceleration sign”. In FIG. 6, the horizontal axis represents the position along the direction perpendicular to the map line segment with the position of the solid white line as the origin, and the vertical axis represents the luminance. Profiles 600 and 610 represent the luminance profiles of the particles to be generated. When the type of map line segment is “white line (thin) + right-side deceleration sign”, not only a single white line but also a broken line, which is a deceleration sign, is marked on the road. Depending on the position in the selected direction, the luminance profile in the direction crossing the map line segment changes depending on the presence or absence of the deceleration marking block. Therefore, in this case, the particle generation unit 23 corresponds to the position of the white line, which corresponds to the case where only the white line exists, represented by the profile 600, with the occurrence probability p1 = 0.4, and the width of the white line. When only the central part with the same width has high luminance and both sides have low luminance, and there is a block of deceleration marking along with the white line represented by the profile 610 at the generation probability p2 = 0.6 The particles having relatively high brightness are generated at two positions corresponding to the white line and the position corresponding to the deceleration marking block. The occurrence probabilities p1 and p2 are set in advance in accordance with the ratio of the length of the deceleration marking block to the length between the blocks.

  FIG. 7 is a diagram illustrating an example of a particle when the type of the map line segment is “white dotted line (thin) + both side deceleration marking”. In FIG. 7, the horizontal axis represents the position along the direction perpendicular to the map line segment with the position of the white dotted line as the origin, and the vertical axis represents the luminance. Profiles 700 to 720 represent luminance profiles of particles to be generated. When the type of map line segment is "white dotted line (thin) + double-sided deceleration sign", not only a single white dotted line but also a broken line as a deceleration sign is marked on the road on both sides of the white dotted line Therefore, depending on the position in the direction along the map line segment, the luminance profile in the direction crossing the map line segment varies depending on the presence or absence of the white dotted line block and the presence or absence of the deceleration marking block. Therefore, in this case, the particle generation unit 23 is high only in the central portion corresponding to the white dotted line, which corresponds to the case where only the white dotted line block represented by the profile 700 exists at the occurrence probability p1 = 0.3. Generate particles that have brightness and low brightness on both sides. Further, the particle generation unit 23 corresponds to a white dotted line block and a deceleration marking block corresponding to a case where there is a deceleration marking block together with the white dotted line block represented by the profile 710 at the occurrence probability p2 = 0.2. Particles with relatively high brightness are generated at three locations. The particle generation unit 23 has high luminance at two places corresponding to the deceleration marking blocks, which corresponds to the case where only the deceleration marking blocks represented by the profile 720 exist at the occurrence probability p3 = 0.5. Generate particles. The occurrence probabilities p1 to p3 are set in advance according to the ratio between the length of the white dotted line block and the length between the blocks, and the ratio between the length of the deceleration marking block and the length between the blocks.

  The particle generation unit 23 passes each generated particle to the projection unit 24.

The projection unit 24 projects each particle onto the image obtained at the current time by the camera 2. If the execution cycle of the vehicle position detection process and the shooting cycle by the camera 2 are not synchronized, the latest image obtained by the camera 2 may be the image obtained at the current time. At that time, the projection unit 24 considers the uncertainty of the predicted position of the vehicle 10 and calculates the predicted position and the predicted direction for each particle according to the normal distribution represented by the covariance matrix when the predicted position is calculated. Correct it. For example, the corrected predicted position and predicted direction μ _q ^ti = (x _q ^ti , z _q ^ti , θ _q ^ti ) are expressed by the following equations.
Here, ε ⁱ represents the correction amount of the predicted position and predicted orientation for the i-th particle, and is obtained by inputting a random number generated for each particle into the normal distribution N (0, Σ _p ^t ).

The projection unit 24 calculates the position of each particle in the vehicle coordinate system based on the corrected predicted position and predicted direction according to the following equation.
Here, g ⁱ represents the position of the i-th particle in the map coordinate system (for example, the position of the end point of the particle or the position where the luminance value changes on the particle), and q ⁱ represents the i-th particle. This represents the position of the particle in the vehicle coordinate system. The rotation matrix R _mv ^ti is a rotation component in the coordinate conversion from the map coordinate system to the vehicle coordinate system at time t, and is determined based on the corrected predicted orientation θ _q ^ti for the i-th particle. Similarly, the parallel _progression ^sequence t _mv ^ti is a translation component in the coordinate transformation from the map coordinate system to the vehicle coordinate system at time t, and the corrected predicted position (x _q ^ti , z _q for the i-th particle). ^ti ).

  The projection unit 24 projects each particle represented by a position on the vehicle coordinate system onto an image obtained by the camera 2 at time t according to the equation (5).

  FIG. 8 is a diagram illustrating an example of an image on which each particle is projected. In the image 800 shown in FIG. 8, a plurality of particles 810 are projected so as to intersect with the white lines 801 to 803 on the road. In this example, the particle generator 23 generates a maximum of 1000 particles for each white line. It can be seen that the position in the direction orthogonal to the white line and the angle formed with the white line are gradually changed by the particles 810. In each particle 810, a relatively white portion indicates that the luminance is relatively low, and a relatively black portion indicates that the luminance is relatively high.

  The projection unit 24 passes each particle projected on the image to the selection unit 25.

  The selection unit 25 calculates an evaluation value representing the degree of coincidence with the image for each particle projected on the image. Then, the selection unit 25 selects particles whose evaluation values satisfy a predetermined matching condition.

  For example, for each particle, the selection unit 25 calculates a normalized cross-correlation value between the luminance value at a predetermined number (for example, 50 points) on the luminance profile of the particle and the luminance value at the corresponding position of the image. Calculated as an evaluation value. Then, the selection unit 25 selects particles whose evaluation value is equal to or greater than a predetermined threshold (for example, 0.8) as satisfying the matching condition. The predetermined threshold value is obtained by, for example, multiplying the maximum evaluation value by a predetermined number less than 1 (for example, 0.8) so that a certain number of particles or more are selected even if the image quality is poor. It may be a value.

  Alternatively, the selection unit 25 may select a predetermined number of particles (for example, 1/5 to 1/10 of the total number of generated particles) in order from the highest evaluation value as satisfying the matching condition.

  Note that the selection unit 25 may calculate an error square sum (SSD) or an error absolute value sum (SAD) as an evaluation value instead of calculating a normalized cross-correlation value for each particle. In this case, the selection unit 25 selects particles whose evaluation value is equal to or less than a predetermined threshold as satisfying the matching condition.

The selection unit 25 estimates a position detection error for each selected particle. The position detection error for the i-th particle includes, for example, an error component due to distortion of the imaging optical system of the camera 2 on the image or an error component due to segmentation in units of pixels, and the covariance matrix R ^Let it be represented by _I ⁱ . When the covariance matrix R _I ⁱ is assumed to increase in a second order as the distance from the center of the image, that is, the optical axis of the camera 2 increases, the covariance matrix R _I ⁱ is expressed by, for example,
Where n ⁱ = (n _u ⁱ , n _v ⁱ ) (n _u ⁱ = (c _u -u ⁱ ) / f _u , n _v ⁱ = (c _v -v ⁱ ) / f _v ) is normalized Represents the center position on the image of the i-th particle in the camera coordinate system. (U ⁱ , v ⁱ ) is the center coordinate on the image of the i th particle, (c _u , c _v ) is the center coordinate on the image, and (f _u , f _v ) Is a parameter representing the focal length of the camera 2. c1 and c2 are constants corresponding to the distortion aberration of the imaging optical system of the camera 2, and are set to c1 = 0.001 and c2 = 0.005, for example. I is a 2-by-2 unit matrix.

The selection unit 25 converts, for each selected particle, the covariance matrix of the detection error on the image into a covariance matrix in the vehicle coordinate system. In this case, the selection unit 25 can calculate the covariance matrix R _q ⁱ at the particle position q ⁱ in the vehicle coordinate system as a bird's-eye transform of the corresponding detection error covariance matrix R _I ⁱ on the image. . This bird's-eye view conversion is expressed by the following equation, for example.
Here, it is assumed that the optical axis direction of the camera 2 is attached so as to be parallel to the road surface. Yc represents the height from the road surface to the camera 2. In this example, components related to the y axis of the vehicle coordinate system that are unrelated to the estimation of the position of the vehicle 10 are omitted. At this time, the covariance matrix R _q ⁱ is expressed by the following equation.
Where B is the Jacobian.

FIG. 9 is a diagram illustrating an example of an error ellipse corresponding to the covariance matrix R _q ⁱ for each selected particle in the vehicle coordinate system. In FIG. 9, the horizontal axis represents the position in the direction orthogonal to the traveling direction of the vehicle 10, and the vertical axis represents the distance from the vehicle 10 along the traveling direction of the vehicle 10. It is assumed that the upper side is farther from the vehicle 10. Each ellipse 901 corresponds to a covariance matrix R _q ⁱ corresponding to one particle. The center point 902 of each ellipse 901 represents the center position of the corresponding particle. As shown in FIG. 9, it can be seen that the error ellipse 901 increases as the particle position moves away from the vehicle 10.

For each selected particle, the selection unit 25 passes the position q ⁱ of the center position of the particle in the vehicle coordinate system, the covariance matrix R _q ^i, and the corresponding map line segment number to the position update unit 26.

  The position update unit 26 estimates the position of the vehicle 10 at the current time based on the estimated position of the vehicle 10 at the most recent past one or more times and each selected particle at the current time.

For example, the position update unit 26 calculates an evaluation value E according to the following equation, and calculates an estimated position of the vehicle at the latest past one or more times and the current time so that the evaluation value E is minimized.
Here, time tε {τ _{s to} τ _e } is a time for estimating the position of the vehicle 10, and τ _e corresponds to the current time t. μ ^t and μ ^t−1 represent the estimated positions of the vehicle 10 at times t and (t−1), respectively. Incidentally, in the calculation of the evaluation value E, for mu ^t-1, the estimated position of the vehicle 10 determined in the preceding time (t-1) is input. That is, μ ^t−1 is treated as a known constant. U ^t is odometry information corresponding to u _t in equation (3). Δ _x (μ ^t , μ ^t−1 , u ^t ) represents an error in predicting the position of the vehicle 10 by odometry, and is expressed by the following expression from the expression (3).
The function g (μ ^t−1 , u ^t ) is expressed by the following equation (3) except that the input variable is the estimated position of the vehicle 10 at time (t−1) and the odometry information at time t. This is the same function as the function g (μ _t−1 , u _t ). That is, the first term on the right side of the equation (11) represents the sum of squares of the error in predicting the position of the vehicle 10 based on the odometry information.

Also, δ _I (μ ^cj , q ^j , χ ^rj ) in the second term on the right side is a time cj (that is, a time corresponding to the selected j-th particle, and any one of the times τ _{s to} τ _e ) At the estimated position μ ^cj of the vehicle 10 for the map line segment r ^j of the jth (j∈ {1,..., I}, I is the total number of particles selected at time τ _{s to} τ _e ). Represents the projection error in the map coordinate system when the particle q ^j is projected onto the map line segment r ^j using the estimated position μ ^cj . That is, the second term represents the sum of squares of the projection error for the selected particle. Note that cj represents the time corresponding to the number j. R _δ ^j represents a covariance matrix of errors in the map coordinate system. The projection error δ _I (μ ^cj , q ^j , χ ^rj ) and the covariance matrix R _δ ^j are expressed by the following equations.
Here, χ _x , χ _z , and χ _c are coefficients and constants when the map line segment in the map coordinate system is expressed by a linear equation, and the linear equation is expressed by the following equation.
X _m and z _m are the coordinate value in the x _m direction and the coordinate value in the z _m direction on the map coordinate system.

The position updating unit 26 calculates the estimated position {μ ^τs ,..., Μ ^τe } of the vehicle 10 at each time so that the evaluation value E is minimized. That is, in the equation (11), the first term μ ^t and the second term μ ^cj are variables. At this time, the position updating unit 26 applies, for example, the gradient method, the Levenberg-Marquardt method, or the Newton method to the equation (11), thereby estimating the estimated position {μ ^{τs,. τe} } can be calculated. Alternatively, the position update unit 26 applies a robust estimation method such as the RANSAC method or the M estimation to the equation (11), and calculates the estimated position {μ ^τs ,..., Μ ^τe } of the vehicle 10 at each time. May be. Thereby, the position update unit 26 can obtain the minimum value of the evaluation value E so that the evaluation value E does not become a local solution by particles having multimodality.

The position update unit 26 stores the estimated position {μ ^τs ,..., Μ ^τe } of the vehicle 10 at each time obtained in the storage unit 41.

  FIG. 10 is an operation flowchart of the vehicle position detection process. The vehicle position detection system 1 estimates the position of the host vehicle at predetermined intervals, for example, at each imaging cycle of the camera 2 according to the following operation flowchart.

  The predicted position calculation unit 22 calculates the predicted position of the host vehicle at the current time t based on the position of the host vehicle at the previous time (t-1) and the odometry information (step S101). Note that the position of the host vehicle at the previous time (t-1) is the initial value of the position of the host vehicle set by the initial value setting unit 21 if the vehicle position detection process performed at the current time t is the first time. . On the other hand, if the vehicle position detection process has already been performed at the previous time (t-1), the position of the host vehicle is obtained by the position update unit 26 at the previous time (t-1).

  The particle generation unit 23 extracts a map line segment included in a predetermined range from the predicted position and included in the shooting range of the camera 2 based on the map information (step S102). Then, for each extracted map line segment, the particle generation unit 23 generates a plurality of particles having a luminance profile in a direction crossing the map line segment according to the type of the map line segment (step S103). ).

  The projection unit 24 projects each particle on the image at the current time obtained by the camera 2 in consideration of the position fluctuation due to the error of the predicted position of the vehicle 10 (step S104).

  For each particle projected on the image, the selection unit 25 calculates an evaluation value indicating the degree of coincidence between the particle and the image, and selects a particle that satisfies the predetermined matching condition (step S105). And the selection part 25 calculates the covariance matrix showing the detection error in a vehicle coordinate system about each of the selected particle (step S106).

  The position update unit 26 calculates the sum of errors between the predicted positions of the vehicle based on the estimated positions at a plurality of times in the past and the odometry information, and the sum of the projection errors of each selected particle calculated based on the detection error. The estimated position of the vehicle at each time is updated so that the evaluation value E based on the sum of the values becomes minimum (step S107). And the control part 43 complete | finishes a vehicle position detection process.

  As described above, the vehicle position detection system can detect the position of the host vehicle without a process of repeated trials like template matching. Therefore, this vehicle position detection system can suppress processing load. In addition, this vehicle position detection system generates a plurality of particles having a luminance profile in the direction crossing the line marked on the road around the vehicle, projects it onto the actually captured image, and matches that image. Since the position of the host vehicle is updated using particles having a good degree, the position of the host vehicle can be accurately detected. Furthermore, since this vehicle position detection system generates particles according to the types of lines marked on the roads around the vehicle with reference to the map information, it matches the lines actually marked on the roads. Probable particles can be generated. Therefore, since this vehicle position detection system can increase the number of particles useful for estimating the vehicle position, it is possible to improve the estimation accuracy of the vehicle position.

  In addition, according to the modification, the particle generation unit 23 may generate particles with different line thicknesses or intervals between the lines for each map line segment that is a particle generation target. This is because the thickness of the lines (white lines, deceleration lines, etc.) marked on the road may vary within the range defined by the road standards, and the orientation of the camera 2 with respect to the lane may cause This is because the thickness of the line may change. For example, the particle generation unit 23 assumes that the variation in the width of the line marked on the road included in the target map line segment follows a normal distribution, and sets the line width or the interval between lines according to the normal distribution. A plurality of changed particles may be generated. Thereby, the particle generation unit 23 can increase the number of particles that coincide with the line marked on the road at the position where the vehicle 10 is actually traveling.

  According to another modification, the particle generation unit 23 may generate a particle for each set of the map line segment located on the left side of the vehicle 10 and the map line segment located on the right side as one set. . As a result, even if the shape of part of the line drawn on the road has changed from its original shape due to repair marks on the road or wear, particles that are unlikely to be selected by mistake Can be generated. In this case, the interval between the map line segment located on the left side of the vehicle 10 and the map line segment located on the right side may change depending on the orientation of the camera 2 with respect to the lane or calibration error. Therefore, similarly to the above-described modification, the particle generation unit 23 assumes that the variation in the interval between the map line segment located on the left side of the vehicle 10 and the map line segment located on the right side follows the normal distribution, and follows the normal distribution. A plurality of particles with different intervals may be generated.

  According to another modification, the particle generation unit 23 may generate planar particles parallel to the road surface of the road.

  FIG. 11 is a diagram showing an example of the relationship between particles generated by this modification and lines marked on the road. In FIG. 11, the line 1100 marked on the road has a sawtooth shape. The vertical axis represents the extending direction of the line 1100 (the position of the vertical axis corresponds to the center of gravity of the line 1100 in the direction orthogonal to the extending direction), and the horizontal axis represents the direction orthogonal to the extending direction of the line 1100. . In this example, a set of three lines having a profile having a luminance change along a direction orthogonal to the extending direction along the extending direction of the line 1100 is generated as one particle 1101 and 1102. In FIG. 11, in the individual lines constituting the particles 1101 and 1102, a relatively bright part has a relatively low luminance, and a relatively dark part has a relatively high luminance. In each line, the portion with high luminance is set according to the corresponding portion of the line 1100. By generating such particles, the vehicle position detection system can appropriately select particles that match the line even when the line marked on the road has a complicated shape. Can be estimated.

  According to still another modified example, when the camera 2 generates a color image, the particle generation unit 23 corresponds to the part of the particle corresponding to the line marked on the road according to the type of the map line segment. May be set according to the color of the line. Alternatively, the particle generation unit 23 may set the color of the part corresponding to the line marked on the road in the particle according to the probability distribution of the color of the line marked on the road.

  According to yet another modification, the camera 2 may be a stereo camera. In this case, the projection unit 24 projects each particle on each of left and right images obtained by, for example, a stereo camera. For example, the selection unit 25 determines a road surface area corresponding to the road surface from each of the left and right images obtained by the stereo camera. Then, the selection unit 25 does not select particles in which some of the particles are not included in the road surface area in either of the left and right images. Thereby, since the selection part 25 can exclude the particle | grains corresponding to the line on the road hidden in the preceding vehicle etc. from estimation of the position of the vehicle 10, the estimation accuracy of the position of the vehicle 10 can be improved more. Note that the selection unit 25 uses, for example, Seki et al., “Recognition of the vehicle's front environment using stereo moving images”, Information Processing Society of Japan Research Report Computer Vision and Image Media (CVIM) 2007, 1- 16, the method disclosed in 2007, etc. can be used.

  In addition, for each particle, the selection unit 25 is the smaller of the evaluation values for the left and right images (when using a normalized cross-correlation value as the evaluation value, when using SSD or SAD as the evaluation value, Or the larger of the evaluation values for the left and right images), or an evaluation value whose average value satisfies a predetermined matching condition may be selected. According to this modification, when an occlusion occurs in which the position where the target particle is projected is hidden by the preceding vehicle, the position on the preceding vehicle that overlaps the projected position is different in the left and right images. Unless there is a pattern similar to the focused particle in two or more places in the preceding vehicle, the particle is not selected. Therefore, robustness is improved.

  Even when the camera 2 is a stereo camera, the position updating unit 26 may calculate the estimated position of the vehicle 10 at each time so that the evaluation value E expressed by the equation (11) is minimized.

  According to still another modification, the particle generation unit 23 may generate particles at the current time based on the particles selected at the immediately preceding time. In this case, the particle generation unit 23 may generate particles at the current time according to a general particle filter. Thereby, the particle generation unit 23 can reduce the number of particles to be generated. That is, the number of particles to be evaluated by the selection unit 25 can be reduced.

  According to still another modification, the position updating unit 26 omits the first term in the expression (11) depending on the required accuracy so that the evaluation value E is minimized based only on the second term. In addition, the position of the vehicle 10 at the current time may be estimated. In this case, the position update unit 26 may calculate the second term based on only the particles selected at the current time also in the second term. Thereby, the position update part 26 can reduce the amount of calculations.

  The position of the host vehicle output from the vehicle position detection system according to the above embodiment or modification is transmitted to a control circuit (not shown) of the driving support system via, for example, CAN3. The control circuit of the driving support system compares, for example, the position of the host vehicle and the surrounding information, and within a predetermined distance range from the host vehicle, for example, a specific structure (for example, a highway toll booth, a route being navigated) If there is an intersection that requires a left turn or a right turn, the driver is notified that the structure is near through a display or a speaker installed in the vehicle. Alternatively, the control circuit of the driving support system may output a command to reduce the speed to the ECU 11.

  As described above, those skilled in the art can make various modifications in accordance with the embodiment to be implemented within the scope of the present invention.

1 Vehicle position detection system 2 Camera 3 Control area network (CAN)
4 Controller (vehicle position detection device)
10 Vehicle 11 ECU
41 storage unit 42 communication unit 43 control unit 21 initial value setting unit 22 predicted position calculation unit 23 particle generation unit 24 projection unit 25 selection unit 26 position update unit

Claims (7)

A storage unit (41) for storing map information indicating the position of the line marked on the road and the type of the line;
A line marked on the road is extracted from the map information in accordance with the predicted position of the vehicle at the current time estimated from the position immediately before the vehicle (10), and for each of the extracted lines, the type of the line A particle generation unit (23) that generates a plurality of particles representing a luminance change in a direction crossing the line while changing a position according to the direction along the line;
For each of the plurality of particles for each of the extracted lines, a position based on the vehicle (10) is obtained while changing the predicted position within an assumed error range, and the vehicle ( A projection unit (24) that projects the image at the current time obtained by the imaging unit (2) mounted on 10);
A selection unit (25) for selecting a particle whose degree of coincidence with the image satisfies a predetermined coincidence condition among the plurality of particles for each of the extracted lines;
A position updater (26) for obtaining the estimated position according to the sum of errors when each of the selected particles is projected onto the corresponding line based on the estimated position of the vehicle (10) at the current time; ,
A vehicle position detection device comprising:   The vehicle position detection according to claim 1, wherein the position update unit (26) obtains the estimated position so that a sum of errors is minimized when each of the selected particles is projected onto the corresponding line. apparatus.   The position update unit (26) calculates odometry information from the total sum of errors when each of the selected particles is projected onto the corresponding line and the estimated position of the vehicle (10) at each of a plurality of past times. Each of the current time and the plurality of past times so that the sum of the error between the position of the vehicle at the next time predicted by using and the estimated position of the vehicle at the next time is minimized. The vehicle position detection device according to claim 1, wherein an estimated position of the vehicle (10) is obtained.   When the extracted line type includes a line including a dotted line or a broken line, the particle generating unit (23) indicates a luminance change in a direction crossing the line corresponding to the case where there is a block of the dotted line or the broken line. The vehicle position as described in any one of Claims 1-3 which produces | generates the 2nd particle showing the brightness | luminance change of the direction crossing the said line corresponding to the case where there is no said dotted line or broken line block Detection device. The imaging unit (2) is a stereo camera including two cameras,
The selection unit (25) is configured so that the degree of coincidence satisfies the coincidence condition on the first image obtained at one time by one of the two cameras and the other of the two cameras The vehicle position detection device according to any one of claims 1 to 4, wherein the coincidence degree selects particles satisfying the coincidence condition on a second image obtained at a current time. Lines marked on the road are extracted from the map information in accordance with the predicted position of the vehicle at the current time estimated from the position immediately before the vehicle (10), and each of the extracted lines is represented in the map information. In accordance with the type of the line, a step of generating a plurality of particles representing a luminance change in the direction across the line while changing the position according to the direction along the line;
For each of the plurality of particles for each of the extracted lines, a position based on the vehicle (10) is obtained while changing the predicted position within an assumed error range, and the vehicle ( Projecting the image at the current time obtained by the imaging unit (2) mounted on 10);
Selecting a particle satisfying a predetermined matching condition with a degree of matching with the image from the plurality of particles for each of the extracted lines;
Obtaining the estimated position according to the sum of errors when each of the selected particles is projected onto the corresponding line based on the estimated position of the vehicle (10) at the current time;
The vehicle position detection method characterized by including. Lines marked on the road are extracted from the map information in accordance with the predicted position of the vehicle at the current time estimated from the position immediately before the vehicle (10), and each of the extracted lines is represented in the map information. In accordance with the type of the line, a step of generating a plurality of particles representing a luminance change in the direction across the line while changing the position according to the direction along the line;
For each of the plurality of particles for each of the extracted lines, a position based on the vehicle (10) is obtained while changing the predicted position within an assumed error range, and the vehicle ( Projecting the image at the current time obtained by the imaging unit (2) mounted on 10);
Selecting a particle satisfying a predetermined matching condition with a degree of matching with the image from the plurality of particles for each of the extracted lines;
Obtaining the estimated position according to the sum of errors when each of the selected particles is projected onto the corresponding line based on the estimated position of the vehicle (10) at the current time;
The computer program for vehicle position detection characterized by including the command which makes the processor (43) mounted in the said vehicle (10) execute.