JP2012167944A - Stereo camera calibration method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a stereo camera that has a plurality of monocular cameras and measures the distance of an object from the parallax of an image picked up by each monocular camera, the stereo camera being configured to improve the calibration accuracy of the stereo camera by accurately measuring the distance from the stereo camera to a calibration target.SOLUTION: The stereo camera comprises two monocular cameras 1 and 2 and a camera stay 3 connecting the monocular cameras. A calibration target 30 is installed in front of the stereo camera and a laser distance member 20 is installed behind the stereo camera. Using the face of the camera stay, which intersects the direction of a calculated distance in perpendicular, as a reference face, the distance L1 of the reference face and the distance L2 of the calibration target are measured by a laser distance meter. Using the fact that the reference face has the known positional relation with the original point of the calculated distance, a distance from the stereo camera to the calibration target is accurately obtained from the measured distances L1 and L2, thereby calibrating a parameter for the stereo camera.

Description

  The present invention relates to a stereo camera used for measuring the distance and position of an object, and more particularly to a method and apparatus for calibrating internal parameters and external parameters of a stereo camera.

  One of the uses of stereo cameras is to acquire images from monocular cameras installed at multiple positions and measure the distance and position of an object using the difference (parallax) between the image positions of each monocular camera. is there.

  In recent years, there has been a growing need for environmental recognition. For example, a system that supports driving by putting a stereo camera in a car and providing information such as the distance between vehicles and the presence or absence of pedestrians to the driver has been put into practical use. It's getting on. For this reason, there is an increasing demand for a long measurement range and high measurement accuracy.

  As described above, in the stereo camera, the distance is obtained from the difference in image position (parallax) between the monocular cameras, but the relationship between the parallax and the distance depends on the parameters of the stereo camera. Stereo camera parameters include internal parameters such as the image center position and focal length of monocular cameras, and external parameters that are relative positions and orientations of monocular cameras.

  Calibration (calibration) is performed in advance to identify the parameters of the stereo camera. The accuracy of distance measurement depends on the accuracy of this calibration. However, since it is difficult to directly measure internal parameters and external parameters, each parameter is generally estimated from a captured image of each monocular camera.

  For example, in Patent Document 1, calibration is performed by photographing a calibration target having known three-dimensional coordinates. In this case, the accuracy of calibration depends on the installation accuracy of the distance between the calibration target and the camera. However, it is difficult to measure the exact positional relationship between the camera and the calibration target.

  The present invention is a stereo camera that includes a plurality of monocular cameras and measures the distance of an object from the difference in image positions of images taken from the monocular cameras, so that the distance from the stereo camera to the calibration target can be accurately measured. The purpose of this is to improve the calibration accuracy of the internal parameters and external parameters of the stereo camera.

The stereo camera of the present invention includes a plurality of monocular cameras and a camera stay that connects the monocular cameras. A calibration target is installed in front of the stereo camera, a distance measuring instrument such as a laser distance meter is installed in the rear, and the distance of the camera stay perpendicular to the calculated distance direction is used as a reference plane. The distance L1 from the distance measuring device to the reference plane and the distance L2 from the distance measuring device to the calibration target are measured. Then, using the fact that the reference plane has a known positional relationship with the origin of the distance to be calculated (optical center of the stereo camera), the distance from the stereo camera to the calibration target is obtained from the measured distances L1 and L2. Calibrate the stereo camera parameters.
Specifically, the distance Z of the object is expressed by the following formula: Z = B * f (d ₀ + Δd)
(B is the baseline length, f is the focal length, d ₀ is the parallax, Δd is the parallax offset),
At two or more positions where the distance from the stereo camera to the calibration target is different, the distance from the stereo camera to the calibration target is measured as described above, and the parallax is calculated by photographing the calibration target with the stereo camera. The Bf value and Δd are calculated by the above formula from a combination of two or more of the measured distance and parallax.

Further, the stereo camera is assumed to be two monocular cameras, and the distance Z of the object is expressed by the following formula: Z = B * f ₀ / d ₀
(B ₀ is the base length of the design value, f ₀ is the focal length of the design value, and d ₀ is the parallax)
Using a calibration target having a plurality of feature points whose three-dimensional coordinates are known, the distance from the stereo camera to the calibration target is measured as described above, and the calibration target is photographed with the stereo camera.
For each monocular camera, the monocular corresponding to the position (a, b, 0) of the feature point on the calibration target in a coordinate system having the center of the calibration target as the origin and the x and y axes in the horizontal and vertical directions, respectively. ideal image position of the feature point in the camera _(i 0, _{j 0)} following equation _{i 0 = (a ± B 0} /2) * f 0 / Z
j ₀ = b * f ₀ / Z
In is calculated, the actual image position of the feature point in the monocular camera (i, j) from the difference between the ideal image position _{(i 0, j 0),} (i, j) of the image reflected in _{(i 0} , J ₀ ) is created.

  According to the present invention, since the distance from the stereo camera to the calibration target can be accurately measured using the distance measuring device, the calibration accuracy of the parameters of the stereo camera is improved, and the distance measurement is performed with high accuracy. It becomes possible.

It is a block diagram of the stereo camera in Example 1 of this invention. FIG. 3 is a diagram illustrating an apparatus configuration example during calibration in the first embodiment. FIG. 3 is a diagram illustrating an overall flow at the time of calibration according to the first embodiment. It is a figure which shows the distance relationship of each part of FIG. FIG. 3 is a block diagram of a processing system at the time of calibration in the first embodiment. FIG. 3 is a block diagram of a processing system during normal use according to the first embodiment. It is a block diagram of the stereo camera in Example 2 of this invention. 6 is a diagram illustrating an example of a device configuration during calibration in Embodiment 2. FIG. FIG. 10 is a diagram illustrating an overall flow at the time of calibration according to the second embodiment. It is a figure which shows the distance relationship of each part of FIG. It is a figure which shows the relationship between the feature point on the target for calibration, and the feature point on a stereo camera. 10 is a block diagram of a processing system at the time of calibration in Embodiment 2. FIG. 6 is a block diagram of a processing system during normal use according to Embodiment 2. FIG.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows the configuration of the stereo camera according to the first embodiment (Example 1) of the present invention. FIG. 1 (a) is an exploded perspective view, and FIG. 1 (b) is a tilted view when completed. ing. The configuration of this stereo camera includes two monocular cameras 1 and 2 and a camera stay 3 to which these monocular cameras 1 and 2 are attached. The monocular cameras 1 and 2 are attached to the camera stay 3 with screws or the like with the front surface of the housing orthogonal to the optical axis as a reference plane. At positions where the monocular cameras 1 and 2 of the camera stay 3 are attached, holes 4 and 5 are provided, respectively. In FIG. 1, x and y are reference planes, and z perpendicular to the reference plane indicates the distance direction.

  Although omitted in FIG. 1, the processing system includes an image processing unit that corrects images captured by the monocular cameras 1 and 2, a parallax calculation unit that obtains parallax using an image output from the image processing unit, and the parallax A distance calculation unit for calculating the object distance based on

  In this embodiment, distortion calibration of each monocular camera 1 and 2 and calibration of external parameters excluding the baseline length are already known methods (for example, Zhang method; “A flexible new technique for camera calibration” IEEE Transactions on Pattern Analysis). and Machine Intelligence, 22 (11): 1330-1334, 2000). The image processing unit corrects images captured by the monocular cameras 1 and 2 based on internal parameters and external parameters obtained by calibration (distortion correction or the like). The parallax calculator calculates the parallax from the captured images of the monocular cameras 1 and 2 corrected by the image processor, and the distance calculator calculates the distance of the object from the parallax.

Here, the distance Z is calculated by equation (1).
Z = B ₀ * f ₀ / d ₀ (1)
B ₀ is the base length of the design value (the length between the optical axes of the camera), f ₀ is the focal length of the design value, and d ₀ is the measured parallax.

  By the way, the actual focal length and baseline length are different from the design values. Further, due to factors such as deviation from the ideal position of the camera sensor (CCD or CMOS sensor), an error also occurs in the measured parallax.

Therefore, in practice, the distance Z is expressed by the following equation: Z = B * f / (d ₀ + Δd) (2)
It becomes. Here, B is the actual baseline length, f is the actual focal length, and Δd is the parallax offset.

  In order to perform distance measurement with a stereo camera with high accuracy, it is necessary to calibrate (calibrate) the BF value (product of B and f) and Δd in advance. In this embodiment, the Bf value and Δd are estimated as follows.

  FIG. 2 shows a configuration example of a calibration apparatus for estimating the Bf value and Δd of the stereo camera of the present embodiment. In FIG. 2, 10 is a stereo camera configured as shown in FIG. 1, 20 is a laser distance meter (distance measuring device), and 30 is a calibration target. In this embodiment, the back surface of the camera stay 3 is a mounting surface for the monocular cameras 1 and 2 and is a reference surface.

  FIG. 3 shows the overall flow of calibration in this embodiment. FIG. 4 shows the distance relationship between the stereo camera 10, the laser distance meter 20, and the calibration target 30.

  First, the mirror 15 is installed on the reference plane which is the back surface of the camera stay 3 constituting the stereo camera 10 (step 101). As shown in FIG. 4, the reflecting surface of the mirror 15 is directed to the rear side of the stereo camera 10. Here, it is assumed that both surfaces of the mirror can be regarded as parallel.

  Next, the laser rangefinder 20 is installed behind the stereo camera 10 so that the laser beam of the laser rangefinder 20 strikes the mirror 15, and the positions of the reference plane and the laser rangefinder 20 are adjusted (step 102). By adjusting the reflected light from the mirror 15 to return to the emission end of the laser distance meter 20, the reference plane of the stereo camera 10 and the laser distance meter 20 are orthogonal to each other, the measurement distance direction of the laser distance meter 20 and the stereo camera Ten measurement distance directions can be matched. Thus, the laser distance meter 20 measures the distance L1 from the laser distance meter 20 to the mirror reflecting surface (step 103). Thereafter, the mirror 15 is removed.

  Next, the calibration target 30 is placed in front of the stereo camera 10 at a position where the laser light of the laser rangefinder 20 is hit (step 104). In this embodiment, since the calibration is performed except for the base line length and the parallax offset, the parallax is constant no matter where the calibration target appears in the screen. Therefore, the calibration target 30 only needs to have one or more feature points that are simultaneously reflected on the monocular cameras 1 and 2. However, in terms of accurately measuring the distance to the feature point, it is desirable to install the laser distance meter 20 so as to be at the laser irradiation position.

  After setting the calibration target 30, the laser distance meter 20 measures the distance L2 from the laser distance meter 20 to the calibration target 30 (step 105). Then, based on the distances L1 and L2, a distance Z from the optical center of the stereo camera 10 to the calibration target 30 is obtained (step 106).

Here, from FIG. 4, the distance Z is
Z = L2-L1-t + ΔL (3)
As required. t is the thickness of the mirror, and ΔL is the distance between the optical center (the origin of distance measurement) and the reference plane, both of which are known.

Next, the calibration target 30 is photographed with the stereo camera 10 (step 107). Then, the parallax d ₀ of the feature point of the calibration target 30 is calculated from the captured images of the monocular cameras 1 and 2 (step 108).

  Thereafter, returning to step 104, the calibration target 30 is set at a position where the distance from the stereo camera 10 is different from the previous time, and steps 105 to 108 are repeated. This is repeated at least for two or more positions where the distance between the stereo camera 10 and the calibration target 30 is different (step 109).

Using two or more sets of the parallax d ₀ and the distance Z obtained in this way, the Bf value and Δd are calculated by the equation (2) (step 110).

  In this embodiment, by providing a reference plane orthogonal to the distance direction of the stereo camera, a distance from the stereo camera to the calibration target can be accurately measured using a laser distance meter (distance measuring device). . Therefore, the estimation accuracy of the Bf value and Δd is improved as calibration of the stereo camera.

FIG. 5 shows a block diagram of a processing system applied during calibration according to the present embodiment. The monocular cameras 1 and 2 of the stereo camera 10 respectively capture the calibration target 30 and output a captured image. The image processing unit 1010 corrects the captured images of the monocular cameras 1 and 2 based on internal parameters and external parameters obtained in advance (distortion correction and the like). The parallax calculation unit 1020 calculates the parallax d ₀ of the calibration target 30 based on the captured images of the monocular cameras 1 and 2 corrected by the image processing unit 1010. For the parallax calculation, for example, a known pattern matching method is used. As described with reference to FIG. 3, the parallax d0 is calculated for each of two or more positions where the distance between the stereo camera 10 and the calibration target 30 is different from each other.

  On the other hand, the distance L1 from the laser distance meter 20 to the mirror reflecting surface and the distance L2 from the laser distance meter 20 to the calibration target are obtained by the laser distance meter 20. The distance L2 is also acquired for two or more positions where the distance between the stereo camera 10 and the calibration target 30 is different from each other.

The calibration calculation unit 1030 receives the parallax d ₀ and the distances L1 and L2, and calculates the Bf value and the parallax offset Δd as follows. It is assumed that the distance ΔL between the optical center of the camera and the reference plane and the mirror thickness t are given in advance.

First, based on L1, L2, ΔL, t, the distance Z from the optical center of the stereo camera 10 to the calibration target 30 is calculated by Equation (3). This is repeated for two or more positions where the distance between the stereo camera 10 and the calibration target 30 is different. The distance Z may be calculated separately in advance. Next, using two or more combinations of the parallax d ₀ and the distance Z, the Bf value and Δd are calculated by Expression (2). Specifically, it is calculated by substituting d ₀ and Z into equation (2) and solving simultaneous equations. The calculated Bf value and Δd may be stored in a memory or the like.

  FIG. 6 shows a block diagram of a processing system applied during normal use of this embodiment. Here, the image processing unit 1010 and the parallax calculation unit 1020 can also be used for calibration.

A subject is photographed by each of the monocular cameras 1 and 2 of the stereo camera 10. The image processing unit 1010 corrects the captured images of the monocular cameras 1 and 2 (distortion correction or the like). Parallax calculation unit 1020 inputs the captured image of the monocular camera 1 that is corrected by the image processing unit 1010, calculates the parallax d ₀ by using the well-known pattern matching method or the like. The distance calculation unit 1040 receives the parallax d ₀ from the parallax calculation unit 1020 and the Bf value and Δd obtained by the previous calibration, and calculates the distance Z of the subject using Expression (2).

  5 and 6, the processing system at the time of calibration and the processing system at the time of normal use are shown separately. However, in practice, these are configured integrally using a CPU, a memory, and the like. Become.

  FIG. 7 shows the configuration of the stereo camera according to the second embodiment (Example 2) of the present invention. FIG. 7 (a) is an exploded perspective view, and FIG. 7 (b) is a tilted view when completed. ing. As in the first embodiment, the stereo camera has two monocular cameras 1 and 2 and a camera stay 3 to which these monocular cameras 1 and 2 are attached. Holes 4 and 5 are provided at positions where the monocular cameras 1 and 2 of the camera stay 3 are attached, respectively, and a reference hole 6 serving as a reference for the stereo camera position is formed at the center of the camera stay 3 with respect to the reference plane It is opened vertically. The two monocular cameras 1 and 2 are mounted with screws or the like symmetrically with respect to the reference hole 6 and at the same height as the reference hole 6.

  The processing system is the same as in the first embodiment, and an image processing unit that performs correction processing on images captured by the monocular cameras 1 and 2 and captured images of the monocular cameras 1 and 2 that are corrected by the image processing unit. A parallax calculation unit that calculates parallax from the distance, a distance calculation unit that calculates the distance of the object based on the calculated parallax, and the like.

  In this embodiment, the calibration of the internal parameters and external parameters of the camera is not performed, and the calibration having a plurality of feature points whose three-dimensional coordinates are known in order to correct the deviation due to the internal parameters or the external parameters in the image processing unit. The correction target lookup table is created by photographing the target.

  FIG. 8 shows a configuration example of the calibration apparatus for the stereo camera of the present embodiment. In FIG. 8, 10 is a stereo camera configured as shown in FIG. 7, 20 is a laser distance meter, and 30 is a calibration target. In the present embodiment, the back surface of the camera stay 3 is a mounting surface for the monocular cameras 1 and 2, but the reference surface is the front surface of the camera stay. The front and back of the camera stay can be regarded as parallel. It is desirable that the diameter of the reference hole 6 on the camera stay 3 is substantially the same as that of the laser beam system of the laser distance meter 20 and is slightly larger. The calibration target 30 is a plate material that can be regarded as a plane on which a grid pattern with known three-dimensional coordinates is drawn. A feature point is provided at the center of the calibration target 30 so that the center position can be seen.

  FIG. 9 shows the overall flow of the calibration in this embodiment. FIG. 10 shows a necessary distance relationship between the stereo camera 10, the laser distance meter 20, and the calibration target 30.

  First, the mirror 15 is installed in the center of the camera stay 3 so that the reflecting surface is in contact with the reference surface (step 201). At this time, the reference hole 6 is blocked by the mirror 15 as shown in FIG.

  Next, the laser distance meter 20 is installed behind the stereo camera 10 (step 202). At this time, the laser beam of the laser rangefinder 20 is adjusted so that the laser beam enters the reflecting surface of the mirror 15 through the reference hole 6 and the reflected light returns to the laser emitting end. The distance meter 20 is made vertical, and the stereo camera 10 and the laser distance meter 20 are aligned. Thus, the laser distance meter 20 measures the distance L1 from the laser distance meter 20 to the mirror reflection surface (reference surface) (step 203). Thereafter, the mirror 15 is removed.

  Next, the calibration target 30 is installed in front of the stereo camera 10 (step 204). At this time, the calibration target 30 is adjusted so that the laser light of the laser rangefinder 20 hits the center of the calibration target 30 through the reference hole 6 of the camera stay 3 and enters the calibration target 30 perpendicularly. . As described above, a pattern (lattice pattern) with known three-dimensional coordinates is drawn on the calibration target 30, and a feature point is provided at the center so that the center position can be seen.

  After setting the calibration target 30, the laser distance meter 20 measures the distance L2 from the laser distance meter 20 to the calibration target 30 (step 205). Then, based on the distances L1 and L2, a distance Z from the optical center of the stereo camera 10 to the calibration target 30 is obtained (step 206).

From FIG. 10, the distance Z is Z = L2−L1 + ΔL (4)
As required. ΔL is the distance between the optical center and the reference surface, and is also known in this embodiment. In this embodiment, it is not necessary to consider the thickness of the mirror 15.

  Next, the calibration target 30 is photographed with the stereo camera 10 (step 207). Then, a lookup table is created from the difference between the image position of the calibration target 30 obtained by photographing and the ideal position (step 208).

Now, consider a coordinate system having the center of the calibration target 30 as the origin and having x and y axes in the horizontal and vertical directions, respectively. In this case, when the feature point (grid point) k on the calibration target 30 is at the position (a, b, 0), the ideal image position (i ₀ , j ₀ ) of the feature point K in the right monocular camera is 11 (a) and 11 (b),
i ₀ = (a−B _0/2 ) * f ₀ / Z (5)
j ₀ = b * f ₀ / Z (6)
It becomes. Also, the ideal position (i ₀ , j ₀ ) of the feature point K in the left monocular camera is also shown in FIGS. 11 (a) and 11 (b).
i ₀ = (a + B _0/2 ) * f ₀ / Z (7)
j ₀ = b * f ₀ / Z (8)
It becomes. Here, B ₀ is the base length of the design value, and f ₀ is the focal length of the design value.

However, the actual image position (i, j) of the feature point k in each of the monocular cameras 1 and 2 of the stereo camera 10 is generally an ideal image position (due to various factors such as focal length deviation, distortion, and baseline length deviation). i ₀ , j ₀ ).

Therefore, a lookup table for correcting the image shown in (i, j) to (i ₀ , j ₀ ) is created. The lookup table is created for each monocular camera 1 and 2 respectively.

  Even in this embodiment, by providing a reference plane orthogonal to the distance direction of the stereo camera, it is possible to accurately measure the distance from the stereo camera to the calibration target using a laser distance meter (distance measuring device). The ideal position can be obtained with high accuracy. Therefore, it is possible to measure the distance of a stereo camera with high accuracy by correcting each monocular camera image to an ideal image based on a lookup table created from the difference between the image position of the calibration target and the ideal image position. Can do.

  FIG. 12 shows a block diagram of a processing system applied at the time of calibration of the present embodiment. The monocular cameras 1 and 2 of the stereo camera 10 respectively shoot the calibration target 30 and output an image. The image processing unit 2010 corrects the captured images of the monocular cameras 1 and 2 based on default internal parameters and external parameters, and outputs a corrected image. The feature point extraction unit 2020 extracts the feature points (grid points) of the calibration target 30 for each monocular camera from the captured images of the monocular cameras 1 and 2 corrected by the image processing unit 2010, and their image positions. (I, j) is output. The feature points are extracted for each lattice point in the calibration target 30 and the image position is output.

The LUT calculation unit 2030 calculates the image position (i, j) corresponding to the feature points (each grid point) of the calibration target 30 from the feature point extraction unit 2020 and the feature points (each grid point) on the calibration target 30. position (a, b), the distance L1, resulting in the laser rangefinder 20 L2, the distance ΔL of the optical center and the reference plane of the camera, by entering the focal length f _0, the following for each monocular cameras 1 In this way, a lookup table is created. The position (a, b) of each lattice point on the calibration target 30 is known.

Here, the right monocular camera will be described. First, based on L1, L2, and ΔL, the distance Z from the optical center of the stereo camera 10 to the calibration target 30 is calculated by Equation (4). The distance Z may be calculated separately in advance. Next, with respect to the position (a, b) of a certain lattice point on the calibration target 30, the ideal image position (i ₀ , j ₀ ) of the lattice point in the right monocular camera is expressed by equations (5), (6 ). Next, the difference between the image position (i, j) of the calibration target 30 and the ideal image position (i ₀ , j ₀ ) is obtained for the lattice point, and the image reflected in (i, j) is determined as (i ₀ , The correction amount is corrected to j ₀ ). This correction amount is obtained for each lattice point of the calibration target 30. Then, based on the correction amount of each grid point, the correction amount of each pixel is calculated by linear interpolation or the like, thereby completing the lookup table for the right monocular camera. Similarly, a lookup table for the right monocular camera is created. The created lookup table is held in a memory or the like.

FIG. 13 shows a block diagram of a processing system applied during normal use of this embodiment. The subjects are photographed by the monocular cameras 1 and 2 of the stereo camera 10, respectively. The image processing unit 2010 corrects the captured images of the monocular cameras 1 and 2 using a lookup table. That is, the image of the subject shown at (i, j) is corrected to (i ₀ , j ₀ ) in units of pixels. Parallax calculation unit 2040 inputs the corrected image of each monocular cameras 1 and 2 which is output from the image processing unit 2010 (ideal image), and calculates the parallax d ₀ by a well-known pattern matching method or the like. The distance calculation unit 2050 receives the parallax d ₀ from the parallax calculation unit 2040, the base length B _{0 of} the design value, and the focal length f ₀ of the design value, and calculates the distance Z of the subject using Expression (1). To do.

  In this embodiment, since the image processing unit 2010 corrects the captured images of the monocular cameras 1 and 2 to ideal images based on the lookup table, the Bf value and the parallax offset Δd are not obtained and the subject image is obtained. The distance Z can be obtained with high accuracy.

  In the present embodiment as well, in practice, the processing systems of FIGS. 12 and 13 are integrally configured using a CPU, a memory, and the like.

  The embodiment of the present invention has been described above, but the present invention is not limited to the illustrated configuration. For example, three or more monocular cameras may be used. In this case, each of the combinations that can be taken by the two monocular cameras may be calibrated and averaged.

1, 2 Monocular camera 3 Camera stay 6 Reference hole 10 Stereo camera 15 Mirror 20 Laser distance meter (distance measuring device)
30 Target for calibration 1010 Image processing unit 1020 Parallax calculation unit 1030 Calibration calculation unit 1040 Distance calculation unit 2010 Image processing unit 2020 Feature point extraction unit 2030 LUT calculation unit 2040 Parallax calculation unit 2050 Distance calculation unit

JP 2006-90756 A

Claims (7)

A plurality of monocular cameras, a camera stay that connects the monocular cameras, an image processing unit that corrects captured images from the monocular cameras, and a parallax calculation unit that calculates parallax from the corrected images from the image processing unit; A method of calibrating parameters of a stereo camera including a distance calculation unit that calculates the distance of an object using the parallax,
A calibration target is installed in front of the stereo camera, a distance measuring device is installed behind the stereo camera,
A distance L1 from the distance measuring device to the reference surface and a distance L2 from the distance measuring device to the calibration target are measured by the distance measuring device using a surface perpendicular to the calculated distance direction of the camera stay as a reference surface. ,
By using the fact that the reference plane has a known positional relationship with the origin of the distance to be calculated, the distance from the stereo camera to the calibration target is obtained from the measured distances L1 and L2, and the parameters of the stereo camera are obtained. Stereo camera calibration method characterized by calibrating The stereo camera calibration method according to claim 1,
A stereo camera calibration method characterized in that a laser distance meter is used as a distance measuring device and the reference surface and the laser light of the laser distance meter are adjusted to be orthogonal to each other. The stereo camera calibration method according to claim 1 or 2,
The distance Z of the object is
Z = B * f (d ₀ + Δd)
Here, B is a baseline length, f is a focal length, d ₀ is a parallax, and Δd is a parallax offset,
The distance from the stereo camera to the calibration target is measured at two or more positions where the distance from the stereo camera to the calibration target is different, and the parallax is calculated by photographing the calibration target with the stereo camera. A stereo camera calibration method, wherein the Bf value and Δd are calculated from the combination of two or more of the above by the above formula. The stereo camera calibration method according to claim 1 or 2,
A stereo camera has two monocular cameras,
The distance Z of the object is
Z = B ₀ * f ₀ / d ₀
Here, B ₀ is the baseline length of the design value, f ₀ is the focal length of the design value, d ₀ is the parallax,
Where
Using a calibration target having a plurality of feature points with known three-dimensional coordinates, measuring the distance from the stereo camera to the calibration target, and photographing the calibration target with the stereo camera,
For each monocular camera, the monocular corresponding to the position (a, b, 0) of the feature point on the calibration target in a coordinate system having the center of the calibration target as the origin and the x and y axes in the horizontal and vertical directions, respectively. ideal image position of the feature point in the camera _(i 0, _{j 0)} following equation _{i 0 = (a ± B 0} /2) * f 0 / Z
j ₀ = b * f ₀ / Z
In is calculated, the actual image position of the feature point in the monocular camera (i, j) from the difference between the ideal image position _{(i 0, j 0),} (i, j) of the image reflected in _{(i 0} , J ₀ ), a lookup table for correction is created. A stereo camera calibration apparatus for performing the stereo camera calibration method according to claim 3,
Means for calculating parallax from images obtained by photographing the calibration target with the stereo camera at two or more positions at different distances from the stereo camera to the calibration target;
Means for calculating the Bf value and Δd by inputting the distance from the stereo camera to the calibration target and the calculated parallax respectively measured at two or more different positions from the stereo camera to the calibration target. When,
Stereo camera calibration device characterized by comprising: The stereo camera calibration device according to claim 5,
The stereo camera calibration apparatus characterized in that the means for calculating the parallax from the image obtained by photographing the calibration target also serves as a parallax calculation unit for calculating the parallax from the corrected image from the image processing unit. A stereo camera calibration apparatus for performing the stereo camera calibration method according to claim 4,
Means for extracting an image position (i, j) of a feature point for each monocular camera from an image obtained by photographing a calibration target with a stereo camera;
For each monocular camera, the extracted image position (i, j) of the feature point on the calibration target, the position (a, b) of the feature point on the calibration target, the focal length f ₀ , and the calibration from the stereo camera By inputting the distance Z to the target, the ideal image position (i ₀ , j ₀ ) of the feature point in the monocular camera is calculated, and the image position (i, j) and the ideal image position (i ₀ , j ₀ ) A means of creating a lookup table from the differences;
Stereo camera calibration device characterized by comprising:

Images