JP2010055340A - Travel path estimation device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve travel path estimation on a road where there exists no road surface marker by highly precisely detecting the boundary between a road and a region outside the road by using height information. <P>SOLUTION: The travel path estimation device is configured to assume that all points in one image in a stereo image exist on a road plane, and to check the similarity between a converted image converted into the way of viewing in one image and an actually input image, and to detect a road shoulder candidate points whose height is different from that of the road surface, and to estimate a travel path by selecting an actual road shoulder point from the relation of the road shoulder candidate marks in two regions arranged on a straight line connecting a retrieval original point with the road shoulder candidate points by using the road shoulder candidate points as a center in a reverse projection road shoulder candidate point image obtained by reversely projecting the obtained road shoulder boundary candidate points to the road surface. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a travel path estimation apparatus and program for estimating a travel path by detecting a boundary between a road and an area outside the road, using a TV camera mounted on the mobile body and imaging the front of the mobile body in the moving direction. It is about.

In order to improve the safety of a vehicle, a technique for estimating a traveling path of the host vehicle from an image of a vehicle-mounted camera has been developed. As a traveling road estimation technique, many techniques for extracting and tracking a road surface marker such as a white line drawn on a road surface from a luminance distribution in an image have been devised (for example, Patent Document 1, Patent Document 2, Patent). Reference 3). A method of detecting a road edge by detecting an edge of a road (for example, Non-Patent Document 1), a method of detecting a road region using color information (for example, Non-Patent Document 2), and the like have been devised. . In addition, a method for detecting a wall on the side of the road from the height information obtained by stereo vision and estimating a boundary with the road (see, for example, Patent Document 4), a straight line toward the vanishing point direction detected in the stereo image A method of projecting on a road surface, associating straight lines with each other on the projected image, and detecting a step on a sidewalk, a guard rail, a side surface of a parked vehicle, etc. (see, for example, Patent Document 5), a stereo image on a road surface A method (for example, Patent Document 6) for detecting a region of a road surface by back projection has been devised.
Japanese Patent No. 3930366 JP 2003-30626 A JP-A-6-119594 JP 2003-20454 A JP 2003-281552 A Japanese Patent No. 2998744 AMWaxman, JJ Lemoigne and LSDavis, “A Visual Navigation System for Autonomous Land Vehicles,” IEEE Journal of Robotics and Automation, vol.RA-3, No.2, pp.124-141, 1987. C. Thorpe, MH Herbert and T. Kanade, “Vision and Navigation for the Carnegie-Mellon Navlab,” IEEE Trans on PAMI, vol. 10, No. 3, pp. 389-392, 1988.

  The road surface marker such as a white line is detected from the image, and the method of estimating the road is the road where the road surface marker does not exist, the road where the road surface marker is fading and disappearing, the road where the road surface marker is hidden by a parked vehicle, etc. Not applicable to In order to estimate the road in these situations, it is necessary to detect the road from features other than the road surface marker. However, with a monocular camera such as a method for detecting the edge of the road or a method for determining the road area by color. In the method of estimating the road area, it is difficult to stably detect the road area having various brightnesses and textures depending on the situation. The method of detecting walls and guardrails from three-dimensional information acquired by stereo vision and estimating the traveling path requires a large calculation cost for searching corresponding points between stereo images, and is a dedicated image for operating in real time. Processing hardware is required. The straight line heading toward the vanishing point detected in the stereo image is projected onto the road surface, and the straight lines are associated with each other on the projected image to detect steps on the sidewalk, guardrails, side surfaces of parked vehicles, etc. Since the correspondence search is performed only in the portion and the height information is obtained, the traveling road can be estimated only at the place where the straight road shoulder exists although the calculation cost is low. The method of detecting a road surface area by back-projecting a stereo image onto the road surface is low in calculation cost, but has a drawback of erroneously detecting reflection or reflection on a wet road surface. Further, due to an error in the image conversion parameter, a road surface marker such as a white line, a pedestrian crossing, a road surface sign, or a contour portion such as a shadow may be erroneously detected.

  The present invention has been made in view of the above, and a traveling road estimation device capable of estimating a traveling road without being affected by road surface texture even on a road where there is no straight road shoulder such as a road surface marker, a guard rail, and a wall surface, and the like The purpose is to provide a program.

  In order to solve the above-described problems and achieve the object, in the present invention, a first and second TV camera that is installed in a vehicle and images the front of the vehicle, and the first TV are provided. A first image memory for storing an image captured by the camera, a second image memory for storing an image captured by the second TV camera, and an image stored in the first image memory are converted into initial conversion parameters. A road for detecting a road shoulder candidate point having a height different from that of the road surface based on the converted image creation unit for conversion using the image, the converted image created by the converted image creation unit and the image stored in the second image memory An outer region detection unit, a back projection road shoulder candidate point image creation unit that creates an image obtained by back projecting the road shoulder candidate points detected by the road outside region detection unit onto the road surface, and the back projection road shoulder candidate point image creation unit. Backprojected shoulder candidate created The boundary points of the road area and the road outside region shall and a traveling path estimation unit for estimating a selected roadway as shoulder point in the image.

  In another aspect of the present invention, the first TV is set using an image input means for inputting an image taken by the first and second TV cameras installed in the vehicle and an initial conversion parameter. Converted image creating means for converting an image input from the camera, and an out-of-road candidate for detecting a road shoulder candidate point having a height different from the road surface from the converted image created by the converted image creating means and the image input from the second TV camera. Created by an area detection means, a backprojection shoulder candidate point image creation means for creating an image obtained by back projecting the road shoulder candidate points detected by the outside road area detection means on the road surface, and the backprojection shoulder candidate point image creation means From the relationship between the number of road shoulder candidate points in the two regions provided on the straight line connecting the search origin and the road shoulder candidate point centered on the road shoulder candidate point from among the road shoulder candidate points in the back-projected road shoulder candidate point image Select shoulder point, the travel path estimating means for estimating the running path, and assumed to function as a.

  As a result, it is assumed that all points in one image in the stereo image are present on the road plane, and the similarity between the converted image converted into the view in the other image and the actually input image is examined. Thus, a road shoulder candidate point having a height different from that of the road surface is detected, and the search origin and road shoulder are centered on the road shoulder candidate point in the back-projected road shoulder candidate point image obtained by back projecting the obtained road shoulder boundary candidate point onto the road surface. A true road shoulder point can be selected from the relationship between the number of road shoulder candidate points in two regions provided on a straight line connecting the candidate points, and the traveling road can be estimated. Therefore, it is possible to estimate the traveling road without being affected by the road surface texture even on a road where there is no straight road shoulder such as a road surface marker, a guard rail, or a wall surface.

  As described above, according to the present invention, it is possible to estimate a traveling road without being influenced by road surface texture even on a road where there is no straight road shoulder such as a road surface marker, a guard rail, or a wall surface.

  Exemplary embodiments of a travel route estimation apparatus and a program according to the present invention will be explained below in detail with reference to the accompanying drawings.

(Embodiment)
FIG. 1 is a block diagram showing a functional configuration of a travel route estimation apparatus 100 according to an embodiment of the present invention. As shown in the figure, this travel path estimation apparatus 100 stores two TV cameras (first and second TV cameras) 1 and 2 and images input from both TV cameras 1 and 2, respectively. First and second image memories 3 and 4, a converted image creation unit 5, an initial conversion parameter storage unit, an outside road area detection unit 6, a back-projected road shoulder candidate point image creation unit 7, and a travel route estimation unit 8 With.

  Both the TV cameras 1 and 2 photograph the front of the vehicle on which the traveling path tracking device of the automobile as a moving body is mounted, and the photographed images are stored in the first and second image memories 3 and 4, respectively. The

  The converted image creation unit 5 creates an image converted into the appearance in the image stored in the second image memory 4 on the assumption that all points in the image stored in the first image memory 3 exist on the road plane. To do.

A method for creating a converted image will be described with reference to FIG. First, the image I ₁ of the first TV camera 1 is backprojected onto the road plane. Then converted backprojected planar regions in the image I _T when captured by the second TV camera 2. The camera coordinate system of the first TV camera 1 is O-XYZ, the camera coordinate system of the second TV camera 2 is O′-X′Y′Z ′, and the focal length is F. Consider a case where a point (x, y) in I ₁ is back-projected to a point P on the road plane, and P is projected to a point (x ′, y ′) in the image plane I ₂ of the camera C ₂ . When the optical axes of the first TV camera 1 and the second TV camera 2 are almost parallel and the measurement range in the Z-axis direction is much wider than the X-axis direction like a road, (x, y) and ( x ′, y ′) holds the following relational expression.

Here, b is a constant, V _y is the y coordinate of the vanishing point, the I _1, I _2, corresponding image coordinates of points existing on a road plane (x, y), (x' , y') Can be obtained by the least square method. For this reason, after the first TV camera 1 and the second TV camera 2 are installed in the vehicle, feature points on the road plane (such as corner points included in the contours of pedestrian crossings and center lines) are interactively associated with each other. The calculated b and V _y are stored in the initial conversion parameter storage unit.

Each point (x, y) of I ₁ of the image coordinates after conversion (x', y ') a calculated from equation (1), creating a transformed image I _T of I _1. The value of x ′ or y ′ after conversion obtained by the expression (1) is within the range of image coordinates of the second TV camera 2 (for example, 0 ≦ x ′ <W in the case of an image having a width W and a height H). , 0 ≦ y ′ <H), the converted image _IT is created using only (x, y) in I ₁ .

Equation (1), since the use of b and V _y in the initial conversion parameter storing section as a fixed value, the conversion error due to shaking of the vehicle body occurs. In the case of paved roads, the fluctuation in the roll direction is small. When pitching occurs, as shown in FIG. 3, a horizontal shift occurs between the converted left image and the right image. The deviation increases in proportion to the vertical coordinates of the image as shown by the arrows in FIG. FIG. 4A shows the result of obtaining the cross-correlation value while shifting one pixel at a time for each scanning line between the converted left image and the right image. A point with a high correlation value is shown brightly. When straight line detection by Hough transform is performed on the image of FIG. 4A, the amount of deviation is represented by a straight line with the maximum number of votes as shown in FIG. This linear expression is
e (y) = βy + γ (2)
Then, the error of the conversion parameter can be corrected using β and γ obtained from this equation. Formula (3) shows a conversion formula obtained by performing pitching correction on Formula (1).

  A converted image is created from the image input to the first image memory 3 by the equation (1), and a cross-correlation value with the image in the second image memory 4 is obtained while shifting one pixel at a time for each scanning line. Next, a linear expression (2) that maximizes the number of votes is obtained by Hough transform, and conversion parameter correction amounts β and γ are obtained. Then, a converted image of the image input to the first image memory 3 is obtained again by Expression (3). With the above processing, it is possible to suppress a deviation between the converted image in the first image memory 3 and the image in the second image memory 4 caused by pitching.

The out-of-road area detection unit 6 checks the luminance similarity for each pixel between the converted image of the image in the first image memory 3 and the image in the second image memory 4, and has an area having a height different from that of the road surface. Is detected as an area outside the road. As the similarity, a normalized luminance difference and an edge direction difference are used in consideration of a brightness difference between stereo images caused by a difference in aperture and sensitivity.
The similarity CorN (x, y) based on the normalized luminance difference is calculated using Equation (4).

l _x and l _y are window sizes used when calculating the similarity.

CorN (x, y) is calculated at each point of I _T and I _{2, and} a binary image I _bn (x, y) (I _bn (x, y) = 255 if CorN with a preset threshold TN (x, y)> TN, I _bn (x, y) = 0 if CorN (x, y) ≦ TN) is obtained.

The degree of similarity CorD (x, y) based on the edge direction difference is calculated by Expression (8). D _T and D ₂ are edge direction images of I _T and I ₂ obtained by the Sobel operator.

l _x and l _y are window sizes for calculating the similarity, as in the case of the normalized luminance difference.

CorD (x, y) is calculated at each point of D _T and D _{2, and} a binary image I _bd (x, y) (I _bd (x, y) = 255 if CorD with a preset threshold value TD (x, y)> TD, I _bd (x, y) = 0 if CorD (x, y) ≦ TD) is obtained.

The outside road area detection unit 6 obtains a logical sum image I _or between the obtained normalized luminance difference image I _bn and the edge direction difference image I _bd, and sets a white pixel area in I _or as a road shoulder candidate point (vehicle Including obstacle areas).

  The back-projected road shoulder candidate point image creation unit 7 back-projects the road shoulder candidate points detected by the out-of-road region detection unit 6 onto the road surface, and bird's-eye image (back-projected road shoulder candidate point image) BV as seen from above the road surface. (See FIG. 6). If the point (x, y) in the image is back-projected to the point (u, v) on the road surface, the following relational expression holds between (x, y) and (u, v).

  Here, RX is a compression (expansion) rate in the X direction, RY is a compression (expansion) rate in the Y direction, and Ylimit is a lower limit value in the image. The coordinates (x, y) in the image have the vanishing point position as the origin.

  The travel path estimation unit 8 searches for a road shoulder in the back projection road shoulder candidate point image BV created by the back projection road shoulder candidate point image creation unit 7. In a rubbed pedestrian crossing or the like, the shoulder area appears densely in the horizontal direction in the backprojected shoulder candidate image BV. Since it is difficult to search for a road shoulder in this dense area, as shown in FIG. 6, a back-projected road shoulder candidate point image BV is projected in the horizontal direction, and an area having a small projection value is searched for a road shoulder point. To do. In FIG. 7, the range of A and B is the search area.

  As shown in FIG. 5, in the back-projected road shoulder candidate point image BV, the obstacle area having a height different from that of the road surface has a radially collapsed shape. Therefore, the road shoulder position is a position where the road shoulder region continuously appears rearward when the back-projected road shoulder candidate point image BV is searched for radially (circle in FIG. 8). The search origin (origin O in FIG. 8) for determining the radial search direction (the direction of the arrow in FIG. 8) on the backprojection image is the boundary line of the out-of-field region (straight line L in FIG. 8) in the backprojection image. And the intersection of the image center line. The search origin is manually calculated from an image input when the first TV camera 1 and the second TV camera 2 are installed in the vehicle.

When a plurality of search areas are set, a search area having a certain vertical width and closest to the vehicle, that is, an area close to the lower end of the backprojection image, is selected, and the start position of the road shoulder search is obtained. In FIG. 7, the search area A is selected. In order to search the left shoulder, only the left half of the back-projected shoulder candidate image BV is targeted. If the vertical width of the search area is too wide, the starting point of the curved shoulder is shifted. Therefore, when the vertical width is equal to or greater than a certain pixel, the area from the bottom of the search area to a certain height is set as the search area. In the selected search area, the parallelogram search window T shown in FIG. 9 is scanned, and the shoulder degree BD of each point is calculated from the equation (14). The search window T is a parallelogram composed of parallel lines parallel to a straight line connecting the road shoulder candidate point and the search origin, and parallel lines in the scanning line direction. For this reason, the search window T has a different shape at each point in the search area. N _W is the number of white spots (shoulder point) in the upper region W of the T, N _B is the number of black points (points on the road surface) in the lower region B of T, N _T is the total number of pixels in the T is there.

  BD shows a high value at the boundary where the shoulder point appears continuously in the radial direction. For this reason, T is scanned in the search area, the value of BD obtained at each point is projected in the vertical direction, and the position P where the projection value is maximum is set as the starting point of the road shoulder search (FIG. 9).

As shown in FIG. 10, each search area A _i is divided in the vertical direction with a fixed width, and a road shoulder position is searched on each divided line L _ij . The search start point position on the lowermost division line is used as the starting point of the road shoulder search, and the road shoulder position on the upper division line is searched in order. In the example of FIG. 10, the shoulder starting point location on the parting line L ₁₁ is the starting point of the search, searches in a similar manner as the shoulder position on the parting line L ₁₂ to determine the starting point of the shoulder search method immediately above . Road shoulder detection position on the L ₁₁ (the road shoulder starting point position) in the center, and scans the L ₁₂ on the range of the left and right constant intervals (in FIG. 10, the range of the double arrow on the parting line), shoulder degree BD Ask for. Then, the position at which the maximum shoulder degree is obtained with the shoulder position in L _12. The same process is sequentially performed on the upper divided lines to obtain the shoulder position on each line. For the non-search area, the lower shoulder position is used as the starting point for the upper shoulder position search. A broken line connecting the road shoulder points obtained on each divided line is output as a traveling road.
(Modification)
In addition, this invention is not limited to the said embodiment, Various deformation | transformation which is illustrated below is possible.
(Modification 1)
In the above embodiment, the road outside region detection unit 6 outputs the obtained road shoulder candidate points as they are to the backprojected road shoulder candidate point image creation unit 7, but the second image memory included in the obtained road shoulder candidate points. At each point of the image I ₂ in FIG. 4, two windows are set up and down as shown in FIG. 12B, and the image (FIG. 12A) and the second image in the first image memory 3 are set in the upper window. The luminance correlation C _T between the image of the memory 4 (FIG. 12B) and the converted image of the image of the second image memory 4 (FIG. 12B) and the image of the first image memory 3 in the lower window. (FIG. 12 (c)) determined the brightness correlation C _B between, C _T and C _B may output to the inverse projection shoulder candidate point image generating section 7 as a road shoulder candidate point a point indicating a large value constant over . For the luminance correlation, a cross-correlation coefficient normalized by the luminance average and variance in the window is used. In obstacle regions such as off-road work piece or vehicle, the value of the luminance correlation C _T is increased, the value of the luminance correlation C _B becomes lower. On the other hand, on the road surface, the value of the luminance correlation C _B increases and the value of the luminance correlation C _T decreases. Only at the boundary point between the obstacle region and the road surface, since C _T and C _B are shown together high value, it is possible to reduce the number of shoulder candidate points in the first modification.
(Modification 2)
In the above embodiment, the back-projected road shoulder candidate point image creation unit 7 back-projects the road shoulder candidate point input from the outside road area detection unit 6 as it is onto the road surface. Candidate points are detected as a continuous sequence of points, so the image was divided into grid blocks, and interpolation processing such as setting all the points in the grid block containing a certain number of road shoulder candidate points to the road shoulder points was performed. Later, it may be back-projected onto the road surface.
(Modification 3)
In the above embodiment, the traveling road search unit detects only one starting point of the road shoulder search, but there is a high possibility that a road shoulder such as a wall surface further exists behind the step or white line in front. Therefore, each search area may be divided into a plurality of areas, road shoulder points may be obtained in each area, and a plurality of broken lines connecting them may be output as a traveling road. For example, it divided into two areas such S _in and S _out as shown in FIG. 11, calculating the position P _i and P _o where the projection value is maximized in each area. When the threshold value or more projection value is preset P _i, the start point of the road shoulder exploring P _i. Further, when the projected value of P _o is equal to or larger than the threshold value and is away from the position of P _{i by} a certain amount, P _o is also set as the starting point of the road shoulder search. Then, as a starting point for both shoulder exploring P _i and P _o, it searches the shoulder points on each division line, and may output the two polygonal lines connecting the shoulder point as the travel path.
(Modification 4)
Further, in the above embodiment, the road search unit searches for a road shoulder point on each divided line from a position close to the host vehicle to a distance, and estimates the road, but in the back-projected road shoulder candidate point image BV The road shoulder degree obtained from the road shoulder candidate points may be stored in a memory having the same size as the back-projected road shoulder candidate point image BV, and the running road may be estimated using a dynamic contour extraction method such as Snakes.
(Modification 5)
In the above embodiment, the case where the present invention is applied to the traveling path estimation apparatus 100 mounted on an automobile has been described. However, the present invention is not limited to this, and an end of an indoor passage or the like is detected. Thus, the present invention can also be applied to the travel route estimation apparatus 100 such as a mobile robot (moving body).
(Modification 6)
Moreover, although the travel route estimation apparatus 100 in the above embodiment may be configured by a hardware circuit having the functions of each unit as shown in FIG. 1, it is described above by causing a computer such as a CPU to execute a program. You may comprise so that each process may be performed. In addition, a program for causing a computer to execute such processing may be provided to the user via a network such as the Internet, or may be provided to the user by being recorded on a recording medium such as a CD-ROM. May be.
(Modification 7)
In the above embodiment, two TV cameras (first and second TV cameras 1 and 2) are provided. However, the present invention is not limited to this, and three or more TV cameras may be provided.

  The present invention relates to a travel path estimation apparatus and program for estimating a travel path by detecting a boundary between a road and an area outside the road, using a TV camera mounted on the mobile body and imaging the front of the mobile body in the moving direction. This is particularly useful when applied to vehicle travel route estimation.

It is a block diagram which shows the function structure of the traveling route estimation apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the image conversion performed by the conversion image creation part of a travel route estimation apparatus. It is a figure for demonstrating the influence which the pitching of a vehicle body has on image conversion in the conversion image preparation part of a travel route estimation apparatus. It is a figure for demonstrating correction | amendment of the parameter by pitching performed by the conversion image creation part of a travel route estimation apparatus. It is a figure for demonstrating the property of the backprojection road shoulder candidate point image produced by the backprojection road shoulder candidate point image creation part of a traveling path estimation apparatus. It is a figure for demonstrating the method of obtaining the area | region which projects a road shoulder candidate point in a horizontal direction, and searches for a road shoulder point in the travel path estimation part of a travel path estimation apparatus. It is a figure which shows an example of the area | region which searches a road shoulder point in the travel route estimation part of a travel route estimation apparatus. It is a figure for demonstrating the method of searching for a road shoulder point in a backprojection road shoulder candidate point image in the travel path estimation part of a travel path estimation apparatus. It is a figure for demonstrating the method explaining the selection method of the area | region for calculating a road shoulder degree, and a road shoulder point in the travel road estimation part of a travel road estimation apparatus. It is a figure for demonstrating the method of searching a road shoulder point sequentially from a search start point in the travel route estimation part of a travel route estimation apparatus. It is a figure for demonstrating the selection method of the road shoulder point at the time of detecting a some traveling path in the traveling path estimation part of a traveling path estimation apparatus. It is a figure which shows an example of two upper and lower windows in the modification 1 of a traveling path estimation apparatus.

Explanation of symbols

A _i search area B area BD provided on a straight line road shoulder W area L _ij provided on a straight line a plurality of horizontal lines 1 first TV camera 2 second TV camera 3 first image memory 4 second image memory 5 conversion Image creation unit 6 Outside road area detection unit 7 Back-projected road shoulder candidate point image creation unit 8 Travel path estimation unit 100 Travel path estimation device

Claims (8)

A first TV camera and a second TV camera mounted on a moving body and capturing an image in front of the moving direction of the moving body;
A first image memory for storing images captured by the first TV camera, and a second image memory for storing images captured by the second TV camera;
A converted image creating unit that converts the image stored in the first image memory using an initial conversion parameter;
Based on the converted image created by the converted image creation unit and the image stored in the second image memory, an outside road area detection unit that detects a road shoulder candidate point having a height different from the road surface;
A back-projected road shoulder candidate point image creating unit that creates an image obtained by back-projecting the road shoulder candidate points detected by the outside road region detecting unit onto the road surface;
A road estimation unit that selects a road shoulder point as a boundary point between the road area and the road outside area in the back projection road candidate point image created by the back projection road candidate point image creation unit; A travel path estimation apparatus characterized by the above.   The travel path estimation unit includes road shoulder candidates in two areas provided on a straight line connecting a search origin and the road shoulder candidate point centered on the road shoulder candidate point at the road shoulder candidate point in the backprojected road shoulder candidate point image. 2. The traveling road estimation apparatus according to claim 1, wherein the traveling road estimation apparatus is configured to select a road shoulder point based on the road shoulder degree obtained from the point relationship and to estimate the traveling road.   The travel path estimation unit sets a search area for searching for a road shoulder point in the backprojected road shoulder candidate point image, and at a road shoulder candidate point in the set search area, the search origin and the road shoulder centering on the road shoulder candidate point. It is configured to estimate the road by selecting a road shoulder point based on the road shoulder degree obtained from the relationship between the number of road shoulder candidate points in two areas provided on a straight line connecting the candidate points. The travel path estimation apparatus according to claim 1.   The traveling path estimation unit sets a search area for searching for a road shoulder point in the backprojected road shoulder candidate point image, divides the set search area by a plurality of horizontal lines, and sets a plurality of divided areas. In each divided region, the road shoulder degree obtained from the relationship between the number of road shoulder candidate points in two regions provided on the straight line connecting the search origin and the road shoulder candidate point with the road shoulder candidate point as the center is projected in the vertical direction, 2. The travel path estimation apparatus according to claim 1, wherein the travel path is estimated using a position having a large projection value as a horizontal position of a road shoulder in the divided area.   5. The road travel estimation unit is configured to sequentially select road shoulder points from a position close to the vehicle to a distance in the back-projected road shoulder candidate point image. 6. Traveling path estimation device.   The travel path estimation unit includes a road shoulder candidate in two areas provided on a straight line connecting a search origin and the road shoulder candidate point centered on the road shoulder candidate point at the road shoulder candidate point in the backprojected road shoulder candidate point image. 2. The traveling road estimation apparatus according to claim 1, wherein the traveling road estimation apparatus is configured to accumulate a road shoulder degree obtained from the relationship of points and to estimate a traveling road by connecting points having a large road shoulder degree.   The travel road estimation device according to any one of claims 1 to 6, wherein the outside road area detection unit is configured to interpolate a road shoulder candidate point having a height different from that of a road surface.   An image input means for inputting images taken by the first and second TV cameras installed in the vehicle, and a converted image creating means for converting the image input from the first TV camera using the initial conversion parameters; The road outside area detecting means for detecting a road shoulder candidate point having a different height from the road surface from the converted image created by the converted image creating means and the image inputted from the second TV camera, and detected by the outside road area detecting means A back-projected road shoulder candidate point image creating means for creating a back-projected image of the road shoulder candidate points on the road surface, and a road shoulder candidate point in the back-projected road shoulder candidate point image created by the back-projected road shoulder candidate point image creating means The road that estimates the road by selecting the true road shoulder point from the relationship between the number of road shoulder candidate points in the two areas provided on the straight line connecting the search origin and the road shoulder candidate point with the road shoulder candidate point as the center Program for causing a constant unit functions as a.