Stereo camera, moving body, and program

Abstract

To provide a stereo camera capable of measuring the distance to an object using only images from two cameras positioned to capture images in different directions.SOLUTION: A stereo camera according to one aspect of the present disclosure includes: two cameras positioned to capture images in different directions and having a common overlapping area in the captured images; a normal image conversion unit that converts each of the captured images from the two cameras into a normal image; an image matching unit that extracts the overlapping area from the two converted images that have been normalized by the normal image conversion unit; a parallax extraction unit that extracts the parallax of an object captured in the overlapping area; and a distance measurement unit that measures the distance to the object by triangulation from the parallax. The normal vectors of the coordinate planes of the two converted images are parallel, and perpendicular to a base line connecting the optical centers of the lenses of the two cameras.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a stereo camera, a moving object, and a program. [Background technology]

[0002] An increasing number of vehicles are equipped with safety systems that ensure safety when driving. These vehicles are equipped with multiple cameras that can provide a panoramic view of the surroundings of the vehicle (see, for example, Japanese Patent Application Laid-Open No. 2017-121058).

[0003] Conventional safety systems also include sensors that use light and sound to measure the distance to objects, and use these sensors to determine whether an object captured in the panoramic image poses a danger to the vehicle. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-121058 Summary of the Invention [Problem to be solved by the invention]

[0005] As described above, conventional safety systems installed in vehicles require a camera to recognize objects and a sensor to measure distance, which increases the number of parts and tends to increase manufacturing costs.

[0006] On the other hand, if images taken by two cameras in the same direction from separate locations are used, it is possible to measure distance using only the camera images. However, since the multiple cameras mounted on a vehicle are positioned to monitor different directions, a new camera must be installed to capture images in the same direction. This results in increased manufacturing costs compared to measuring distance using a sensor.

[0007] The present disclosure has been made in light of the above circumstances, and aims to provide a stereo camera that can measure the distance to an object using only images from two cameras that are positioned to capture images in different directions, a program, and a moving body that uses this stereo camera. [Means for solving the problem]

[0008] A stereo camera according to one embodiment of the present disclosure includes two cameras arranged to capture images in different directions and having a common overlapping area in the captured images; a normal image conversion unit that converts each of the captured images of the two cameras into a normal image; an image matching unit that extracts the overlapping area from the two converted images that have been normalized by the normal image conversion unit; a parallax extraction unit that extracts the parallax of an object captured in the overlapping area; and a distance measurement unit that measures the distance to the object by triangulation from the parallax, wherein the normal vectors of the coordinate planes of the two converted images are parallel and perpendicular to a base line connecting the optical centers of the lenses of the two cameras.

[0009] A moving object according to another aspect of the present disclosure includes the stereo camera of the present disclosure, and one of the two cameras is positioned so as to capture an image in the forward direction.

[0010] A moving object according to yet another aspect of the present disclosure includes the stereo camera of the present disclosure and moves toward the overlap region.

[0011] Another aspect of the present disclosure provides a program for causing a computer to measure the distance to an object photographed in the overlapping area from images captured by two cameras arranged to capture images in different directions, the captured images having a common overlapping area, the normal vectors of the coordinate planes of the two orthogonally transformed transformed images being parallel and perpendicular to a baseline connecting the optical centers of the lenses, the program comprising: an orthogonal transformation step for orthogonally transforming the images captured by the two cameras; an image matching step for extracting the overlapping area from the two transformed images orthogonally transformed in the orthogonal transformation step; a parallax extraction step for extracting the parallax of the object; and a distance measurement step for measuring the distance to the object from the parallax by triangulation. [Effects of the Invention]

[0012] The stereo camera and program disclosed herein can measure the distance to an object using only images from two cameras positioned to capture images in different directions. The moving object disclosed herein is equipped with the stereo camera disclosed herein, and can measure the distance to the object using the camera that recognizes the object, thereby reducing the number of parts and manufacturing costs. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 is a schematic plan view showing a moving body (automobile) according to an embodiment of the present disclosure. [Figure 2] FIG. 2 is a schematic block diagram showing the configuration of the stereo camera of FIG. [Figure 3] FIG. 3 is an explanatory diagram for explaining the processing of the normal image conversion unit of the stereo camera of FIG. [Figure 4] FIG. 4 shows an example of an image taken by each camera of the stereo camera shown in FIG. [Figure 5] FIG. 5 shows converted images obtained by converting the images in FIG. 4 into normal images. [Figure 6] FIG. 6 is an explanatory diagram for explaining the processing of the distance measurement unit of the stereo camera of FIG. DETAILED DESCRIPTION OF THE INVENTION

[0014] [Description of the embodiments of the present disclosure] First, embodiments of the present disclosure will be listed and described.

[0015] (1) A stereo camera according to one embodiment of the present disclosure includes two cameras arranged to capture images in different directions and having a common overlapping area in the captured images; a normal image conversion unit that converts the captured images of the two cameras into normal images; an image matching unit that extracts the overlapping area from the two converted images that have been normalized by the normal image conversion unit; a parallax extraction unit that extracts the parallax of an object captured in the overlapping area; and a distance measurement unit that measures the distance to the object by triangulation from the parallax, wherein the normal vectors of the coordinate planes of the two converted images are parallel and perpendicular to a base line connecting the optical centers of the lenses of the two cameras.

[0016] In this stereo camera, the two cameras are positioned to capture images in different directions. Therefore, the stereo camera performs orthogonal image transformation so that the normal vectors of the coordinate planes of the transformed images are parallel and perpendicular to the baseline. Therefore, the transformed images obtained from the two cameras can be treated as if they were captured with normal lenses facing the same direction. Therefore, it is possible to measure the distance to an object by triangulation from the parallax between these two transformed images. Note that the normal vector refers to a vector in the actual space (real space) on the side of the object being captured. Furthermore, the same direction in which the normal lens faces is the direction of the optical axis, which coincides with the direction of the normal vector.

[0017] (2) In the stereo camera described in (1) above, it is preferable that the viewing angles of the two cameras are both 120 degrees or more. By using a camera with a wide viewing angle, it is possible to increase the area of ​​the overlapping region common to the images captured by the two cameras, thereby increasing the number of objects whose distances can be measured.

[0018] (3) In the stereo camera described in (2) above, the lenses of the two cameras may be fisheye lenses. Fisheye lenses produce significant distortion, especially at the edges of the image, but they have a wide viewing angle and can easily be converted into a normal image by tilting the normal vector.

[0019] (4) In any of the stereo cameras described in (1) to (3), an optimization unit may be provided that determines the direction of the normal vector so as to maximize the area of ​​the overlapping region. By providing such an optimization unit, it is possible to increase the number of objects whose distances can be measured.

[0020] (5) In the stereo camera according to any one of (1) to (4) above, the two cameras may be movable cameras. By using such movable cameras, distance measurement can be performed with higher accuracy.

[0021] (6) A moving body according to another aspect of the present disclosure is equipped with any one of the stereo cameras described in (1) to (4) above, and one of the two cameras is positioned to capture the forward direction.

[0022] In a mobile object equipped with multiple cameras for monitoring its surroundings, the stereo camera of the present disclosure can be configured using two of the multiple cameras, eliminating the need to install additional components such as cameras. Even in a mobile object without cameras, the stereo camera of the present disclosure can be installed to monitor its surroundings. Furthermore, by positioning one of the two cameras to capture images in the forward direction, it becomes possible to measure the distance to an object in a direction that decreases over time, which is effective in safely controlling the mobile object.

[0023] (7) In the moving body described in (6) above, the other camera may be positioned so as to capture an image in a direction perpendicular to the forward direction. By positioning the other camera in this manner, it is possible to monitor a wide area with a small number of cameras while ensuring an overlapping area.

[0024] (8) A moving object according to yet another aspect of the present disclosure includes the stereo camera of the present disclosure and moves toward the overlapping region.

[0025] In a mobile object equipped with multiple cameras for monitoring its surroundings, the stereo camera of the present disclosure can be configured using two of the multiple cameras, eliminating the need to install additional components such as cameras. Even in a mobile object without cameras, the stereo camera of the present disclosure can be installed to monitor its surroundings. Furthermore, by configuring the mobile object to move toward the overlapping area, it becomes possible to measure the distance to an object in the forward direction, which is effective in safely controlling the mobile object.

[0026] (9) Another aspect of the present disclosure provides a program for causing a computer to measure the distance to an object photographed in the overlapping area from images captured by two cameras arranged to capture images in different directions, the captured images having a common overlapping area, the normal vectors of the coordinate planes of the two orthogonally transformed transformed images being parallel and perpendicular to a baseline connecting the optical centers of the lenses, the program comprising: an orthogonal transformation step for orthogonally transforming the images captured by the two cameras; an image matching step for extracting the overlapping area from the two transformed images orthogonally transformed in the orthogonal transformation step; a parallax extraction step for extracting the parallax of the object; and a distance measurement step for measuring the distance to the object from the parallax by triangulation.

[0027] This program uses two cameras positioned to capture images in different directions, and performs orthogonal image transformation so that the normal vectors of the coordinate planes of the converted images are parallel and perpendicular to the baseline. Therefore, the converted images obtained from the two cameras can be treated as if they were taken with normal lenses facing the same direction. Therefore, it is possible to measure the distance to an object by triangulation using the parallax between these two converted images.

[0028] [Details of the embodiments of the present disclosure] A stereo camera and a moving object according to an embodiment of the present disclosure will be described below.

[0029] A moving object 100 shown in FIG. 1 is an automobile, and is equipped with a stereo camera 1 according to another embodiment of the present disclosure.

[0030] 2, the stereo camera 1 includes two cameras 10 (a first camera 11 and a second camera 12), an orthogonal image conversion unit 20, an image matching unit 30, an optimization unit 40, a parallax extraction unit 50, and a distance measurement unit 60. Of these, the orthogonal image conversion unit 20, the image matching unit 30, the optimization unit 40, the parallax extraction unit 50, and the distance measurement unit 60 can be implemented using, for example, a microcontroller, such as an in-vehicle microcomputer.

[0031] <Camera> The two cameras 10 are positioned to capture images in different directions, and the captured images have a common overlapping area S. Specifically, one of the two cameras 10, the first camera 11, is positioned to capture images in the forward direction X. The other camera, the second camera 12, is positioned to capture images in the left side direction Y, which is a direction perpendicular to the forward direction X. By positioning the second camera 12 in this manner, it is possible to monitor a wide area with fewer cameras while ensuring the overlapping area S.

[0032] The lower limit of the separation distance between the two cameras 10 is preferably 10 cm, and more preferably 20 cm. On the other hand, the upper limit of the separation distance between the two cameras 10 is preferably 200 cm, and more preferably 100 cm. In the stereo camera 1, the separation distance between the two cameras 10 is used as the base line length, and the distance to the object photographed in the overlapping area S is measured by triangulation. Therefore, if the separation distance between the two cameras 10 is less than the above lower limit, the accuracy of the measured distance may be insufficient. Conversely, if the separation distance between the two cameras 10 exceeds the above upper limit, it may be difficult to arrange the cameras on the moving body 100. Note that the "separation distance" refers to the distance between the optical centers of the lenses of the two cameras.

[0033] The lower limit of the viewing angle of each of the two cameras 10 is preferably 120 degrees, more preferably 150 degrees, and even more preferably 180 degrees. Meanwhile, the upper limit of the viewing angle is preferably 220 degrees, more preferably 200 degrees. If the viewing angle is below the lower limit, the overlapping region S may become narrow, potentially limiting the objects whose distances can be measured. Conversely, if the viewing angle exceeds the upper limit, it may become difficult to achieve high-quality image correction in the normal image conversion unit 20 (described later), or the field of view may overlap with a blind spot of the moving object 100, resulting in an area where a valid image cannot be captured. While the viewing angles of the two cameras 10 may be different, it is preferable that they are equal for ease of processing. Furthermore, it is more preferable that the two cameras 10 are the same type of camera.

[0034] The lenses of the two cameras 10 are preferably fisheye lenses. Fisheye lenses produce significant distortion, especially at the edges of the image, but they have a wide viewing angle and can easily be converted into a normal image by tilting the normal vector n.

[0035] The following description will be given taking as an example a case where the lenses of the two cameras 10 are fisheye lenses with a viewing angle of 180 degrees, but this does not mean that the lenses of the two cameras 10 are limited to this configuration.

[0036] <Normal image conversion section> The normal image conversion unit 20 converts the images captured by the two cameras 10 into normal images.

[0037] When a fisheye lens is used, for example, by using an equidistant projection method, all subjects are projected onto a hemisphere of radius R at positions where the angles α and β in FIG. 3 are equal, as shown in FIG. 3. Furthermore, each of the projected coordinates is mapped onto the XY plane to form a captured image. An example of an image captured by two cameras 10 is shown in FIG. 4.

[0038] As shown in Fig. 3, the normal image conversion unit 20 converts the vector n into an image on the uv plane tangent to a hemisphere of radius R, using the vector n as a normal vector, to obtain a normal image. At this time, normal vectors n1 and n2 in the coordinate plane (uv plane) of the two converted images in real space (XYZ space) are parallel, as shown in Fig. 2. In addition, the normal vectors n1 and n2 are perpendicular to the base line L connecting the optical centers of the lenses of the two cameras 10.

[0039] There are various known methods for the above-mentioned image conversion. For example, when using a fisheye lens with an orthogonal projection method, the following formula (1) can be used, where the magnification of the camera 10 is m and the other variables are as shown in Figure 3.

number

[0040] The projection method is determined by the characteristics of the lens used, but is not limited to orthogonal projection. Image conversion is also possible using lenses designed with projection methods other than orthogonal projection. Image conversion can be performed using a theoretical formula corresponding to the projection method, as in the above-mentioned orthogonal projection method. Alternatively, image conversion can be performed using actual lens design data, i.e., the relationship between the angle of incidence and the image height. Furthermore, by detecting distortion in an image captured using the target lens, conversion can be performed using a conversion coefficient to improve accuracy.

[0041] For example, in the case of equidistant projection, which is the most commonly used type of fisheye lens, if the angle between the light ray that has left the object (the light ray that passes through the coordinates of interest on the uv plane) and the optical axis of the fisheye lens is θ, then the coordinates (x d , y d ) can be determined by the following equation (2) for the coordinates (X, Y) obtained by the above-mentioned orthogonal projection method.

number

[0042] <Image Matching Section> The image matching unit 30 extracts an overlapping region S from the two converted images that have been normalized by the normal image conversion unit 20.

[0043] The image matching unit 30 can be configured, for example, by a preprocessing step, a rectification step, and a block matching step.

[0044] (Pre-processing step) The two converted images are normalized so that they are not observed at different brightnesses even when viewing the same area due to differences in the hardware parameters of the camera 10 or the angle of the line of sight. Specifically, the average value of the brightness values ​​is calculated for each converted image, and this average value is subtracted from each brightness value. In other words, normalization is performed by taking the brightness values ​​as the difference from the average value.

[0045] (Parallelization step) As described above, all objects are projected onto points on a hemisphere of radius R. However, objects located on a line connecting the center of the camera 10's lens (the origin of the XYZ space in Figure 3) and a point on the hemisphere are projected onto the same point. In two transformed images, this line projected onto the same point in one transformed image is not projected onto a single point in the other transformed image due to parallax, and is actually observed as a straight line. This line is called an epipolar line. Unless the two cameras 10 are aligned strictly horizontally, the epipolar line will not be horizontal but will extend diagonally. However, if the epipolar line extends diagonally, the efficiency of image matching calculations decreases. By converting the image to create a normal image using the normal image conversion unit 20 of the stereo camera 1, it is possible to position the epipolar line horizontally. However, fluctuations such as individual differences in the lens and errors in installation position may affect the horizontality of the epipolar line. This can be fine-tuned by rectification based on the actual image, and this process is performed in the rectification step. This rectification step is also called rectification.

[0046] (Block Matching) In the block matching step, a method for calculating the correlation value of the images is used to search for corresponding parts in the two transformed images. An example is shown in Figure 5. In Figure 5, parts (feature points) that have been searched for and determined to correspond in the two transformed images on the left and right are shown with two-dot dashed lines, and the corresponding parts are shown with dashed lines. Figure 5 shows three feature points as an example, but in reality, many more feature points are extracted. This makes it possible to identify the overlapping area S.

[0047] <Optimization section> The optimization unit 40 determines the directions of the normal vectors n1 and n2 so as to maximize the area of ​​the overlap region S. By providing such an optimization unit 40, it is possible to increase the number of objects for which distance can be measured. The optimization unit 40 can be implemented, for example, by the following method.

[0048] First, set the initial values ​​of α and β in Fig. 3. These initial values ​​can be, for example, the direction of the bisector of the angle formed by the different directions in which the two cameras 10 are taking pictures.

[0049] The scanning range for varying α and β is determined in advance. The maximum range that can be recognized as the overlapping region S can be identified from the difference in the viewing angle, mounting position, and shooting direction angle of the camera 10, so the ranges of α and β can be determined so that the normal vector n passes through this range, for example.

[0050] According to the above-mentioned initial values ​​and scan range, the image matching unit 30 is repeatedly executed, and α and β are determined so that the number of points found by a known algorithm such as edge detection that detects feature points in the image is maximized in the extracted image matching unit 30.

[0051] <Parallax extraction part> The parallax extraction unit 50 extracts the parallax of the objects photographed in the overlapping region S. The corresponding parts extracted by the image matching unit 30 form different angles with the normal vectors n1 and n2 of each image, and are observed at different positions on the two images. This difference (distance) in positions on the images is called "parallax."

[0052] After extracting the disparity, the disparity extraction unit 50 may perform an outlier removal step and a hole filling step.

[0053] (Outlier removal step) In the outlier removal step, the disparity values ​​of the two images are checked to see if they are consistent. If there is a specific disparity mainly due to occlusion, that disparity is removed.

[0054] (Fill-in-the-blank step) In the hole filling step, parts that have been removed as outliers and parts at the edge of the screen where the parallax cannot be estimated are filled. Basically, holes are filled using the parallax value on the side with smaller parallax around the removed pixels, i.e., the background side. Note that it is also possible to only remove parallax and omit the hole filling step.

[0055] <Distance measurement section> The distance measurement unit 60 measures the distance to the object by triangulation from the parallax.

[0056] As shown in FIG. 6, the parallax is calculated by the similarity of a triangle using the distance Z from the lens of the camera 10 to the object, the focal length F of the lens of the camera 10, the base length B, and the parallax D. Z=B×F / D (3) It can be calculated as follows.

[0057] <Program> The operations of the orthogonal image conversion unit 20, image matching unit 30, optimization unit 40, parallax extraction unit 50, and distance measurement unit 60 of the stereo camera 1 can be performed by a program. That is, a program according to another embodiment of the present disclosure is a program for causing a computer to execute distance measurement to an object photographed in the overlapping area S from images photographed by two cameras 10, the cameras 10 being arranged to photograph in different directions, the photographed images having a common overlapping area S, and the normal vectors n1, n2 of the coordinate planes of the two orthogonally converted transformed images being parallel and perpendicular to a base line L connecting the optical centers of the lenses, the program including an orthogonal image conversion step of respectively converting the photographed images of the two cameras 10 into orthogonal images, an image matching step of extracting the overlapping area S from the two transformed images orthogonally converted in the orthogonal image conversion step, a parallax extraction step of extracting the parallax of the object, and a distance measurement step of measuring the distance to the object by triangulation from the parallax.

[0058] <Advantages> In the stereo camera 1, the two cameras 10 are positioned so as to capture images in different directions. Therefore, the stereo camera 1 performs orthogonal image transformation so that the normal vectors n1 and n2 of the coordinate plane of the transformed images are parallel and perpendicular to the base line L. Therefore, the transformed images obtained from the two cameras 10 can be treated as if they were captured by normal lenses facing in the same direction. Therefore, it is possible to measure the distance to an object by triangulation from the parallax of these two transformed images.

[0059] In a mobile object 100 equipped with multiple cameras 10 for monitoring its surroundings, the stereo camera 1 of the present disclosure can be configured using two of the multiple cameras 10, eliminating the need to install additional components such as cameras 10. In a mobile object 100 that does not have any cameras 10, the stereo camera 1 of the present disclosure can be installed to monitor its surroundings. Furthermore, by positioning the first camera 11 of the two cameras 10 so that it captures images in the forward direction X, it becomes possible to measure the distance to an object in a direction that decreases over time, which is effective in safely controlling the mobile object 100.

[0060] The program also performs orthogonal image transformation using two cameras 10 positioned to capture images in different directions so that the normal vectors n1 and n2 of the coordinate plane of the converted images are parallel and perpendicular to the base line L. Therefore, the converted images obtained from the two cameras 10 can be treated as if they were captured using normal lenses facing the same direction. This makes it possible to measure the distance to an object by triangulation from the parallax between these two converted images.

[0061] [Other embodiments] The above-described embodiments do not limit the configuration of the present disclosure. Therefore, the above-described embodiments may include omissions, substitutions, or additions of components based on the description in this specification and common general technical knowledge, and all of these should be construed as belonging to the scope of the present disclosure.

[0062] In the above embodiment, an example was described in which two cameras capture images in the forward direction and in the left side direction perpendicular to the forward direction, but this configuration is not limited to this as long as the capture directions of the two cameras are different.

[0063] The upper limit of the angle between the shooting directions of the two cameras is preferably 90 degrees, more preferably 60 degrees, and even more preferably 30 degrees. If the angle exceeds the upper limit, the overlapping area common to the images captured by the two cameras will become narrow, which may limit the objects to which distances can be measured. On the other hand, the lower limit of the angle is not particularly limited and may be 0 degrees.

[0064] The two cameras can also be movable cameras, which allows for more accurate distance measurement.

[0065] In the above embodiment, the moving object is described as having two cameras, but this does not exclude the moving object from having other cameras, and the stereo camera of the present disclosure may be configured with the other cameras. For example, the moving object 100 shown in FIG. 1 may be provided with a third camera (not shown) that captures images in the right side direction, and the stereo camera of the present disclosure may be configured using the first camera 11 and the third camera.

[0066] In the above embodiment, the stereo camera is described as including an optimization unit, but the optimization unit is not an essential component and can be omitted. When the optimization unit is omitted, the normal vector of the coordinate plane of the converted image can be set to a pre-fixed direction, for example, the direction of the bisector of the angle formed by the different directions in which the two cameras are photographing.

[0067] In the above embodiment, the moving body is a vehicle, but the moving body is not limited to a vehicle. For example, the moving body may be a flying drone or a transport robot.

[0068] If the moving body is a flying robot, unlike a vehicle, it moves in three-dimensional space, and therefore it is preferable to have a camera that captures images not only in the horizontal plane but also in the vertical direction. Furthermore, the moving body (flying robot) can be configured to move toward the overlapping area. By configuring the moving body to move toward the overlapping area, it becomes possible to measure the distance to an object in the forward direction, which is effective in safely controlling the moving body.

[0069] Furthermore, when the moving body is a transport robot, monitoring the front, which corresponds to the transport direction, is particularly important, so it is preferable to provide a camera that can capture images in the transport direction, the left and right lateral directions, as well as diagonally forward left and right directions, and particularly in the case of outdoor transport, a downward forward direction. [Industrial Applicability]

[0070] The stereo camera of the present disclosure can measure the distance to an object using only images from two cameras positioned to capture images in different directions. Since the moving body of the present disclosure is equipped with the stereo camera of the present disclosure, the distance to the object can be measured using the camera that recognizes the object, thereby reducing the number of parts and manufacturing costs. [Explanation of symbols]

[0071] 1 Stereo camera 10 Camera 11 Camera 1 12 Second Camera 20. Normal image conversion unit 30 Image matching section 40 Optimization Section 50 Parallax extraction part 60 Distance measurement unit 100 Mobile X Forward direction Y Left side direction S overlap area L baseline n, n1, n2 normal vectors

Claims

1. two cameras positioned to capture images in different directions, the captured images having a common overlapping region; a normal image conversion unit that converts each of the images captured by the two cameras into a normal image; an image matching unit that extracts the overlapping region from the two converted images that have been normalized by the normal image conversion unit; a parallax extraction unit that extracts parallax of an object photographed in the overlapping area; a distance measurement unit that measures the distance to the object by triangulation from the parallax; Equipped with A stereo camera in which the normal vectors of the coordinate planes of the two transformed images are parallel and perpendicular to a base line connecting the optical centers of the lenses of the two cameras.

2. 2. The stereo camera according to claim 1, wherein the viewing angles of the two cameras are both 120 degrees or more.

3. 3. The stereo camera according to claim 2, wherein the lenses of the two cameras are fisheye lenses.

4. The stereo camera according to claim 1 , further comprising an optimization unit that determines the direction of the normal vector so as to maximize the area of ​​the overlapping region.

5. 4. The stereo camera according to claim 1, wherein the two cameras are movable cameras.

6. A stereo camera according to any one of claims 1 to 3 is provided, A moving body in which one of the two cameras is positioned so as to capture images in the forward direction.

7. 7. The moving body according to claim 6, wherein the other camera is positioned so as to capture an image in a direction perpendicular to the forward direction.

8. A stereo camera according to any one of claims 1 to 3 is provided, A mobile object moving toward the overlapping region.

9. A program for causing a computer to execute distance measurement from images taken by two cameras arranged to take images in different directions, having a common overlapping area in the captured images, with normal vectors of coordinate planes of two orthogonally transformed converted images being parallel and perpendicular to a base line connecting the optical centers of the lenses, to an object captured in the overlapping area, a normal image conversion step of converting the images captured by the two cameras into normal images; an image matching step of extracting the overlapping region from the two transformed images that have been normalized in the normal image transformation step; a parallax extraction step of extracting the parallax of the object; a distance measurement step of measuring a distance to the object by triangulation from the parallax; A program that includes:

Images