JP5311016B2 - Stereo camera unit and stereo matching method - Google Patents

Description

  The present invention relates to a stereo camera unit and a stereo matching method.

  The stereo matching method uses two sets of cameras arranged on the left and right, and uses a set of images taken with two cameras to correlate the correspondence between the image from one camera and the image from the other camera. This is a known method for measuring the position, distance, etc. of each point of an object based on the principle of triangulation from the parallax obtained from the corresponding relationship.

Patent Document 1 discloses a basic stereo matching method using a pyramid image, and in order to provide a correlation calculation method for calculating a distance distribution to an object at high speed, the object is imaged with a pair of imaging devices, After making the image into a multi-resolution image with a pyramid structure, stereo matching is performed between each layer. At this time, first, it is performed between upper-layer images that are low-resolution images, and an approximate parallax is obtained. The stereo matching is performed up to the final layer by repeating the setting of the matching region and the search region for the stereo matching. Such a correlation calculation method is applicable to a distance distribution detection device that calculates a distance distribution to an object mounted on a vehicle and surrounding the vehicle, for example.
Japanese Patent Laid-Open No. 2001-319229 (Summary, see FIGS. 1 and 3)

  However, when a stereo camera system that requires a large parallax in distance measurement is configured according to the above-described conventional technology, it is necessary to increase the number of pyramid stages in a pyramid structure including multi-resolution images. If it does so, the frequency | count of a calculation will increase by the number of pyramid steps | stages, and will interfere with a high-speed process.

  An object of the present invention is to provide a stereo camera unit and a stereo matching method capable of high-speed processing of correlation calculation in stereo matching in view of the above-described problems of the conventional technology.

In order to achieve the above object, the stereo camera unit according to the present embodiment includes a first camera that captures a reference image, each center position of which is on the same base line, and the first camera. A second camera that captures a first reference image provided at a position having a predetermined reference line length L, and a position having a predetermined reference line length k × L (k> 1) with respect to the first camera. a third camera for photographing the second reference image provided in the front Kimoto quasi image and a window to the first reference image sets respectively, performed by the phase-only correlation method correlation calculated by the two window a preliminary correlation execution unit for calculating a first parallax ds (u, v) by the first disparity ds (u, v) from the preliminary correlation performing unit to k multiplied by k × ds (u, v) setting a window of the second reference image based on the parallax , And a correlation execution unit for executing the phase-only correlation method a correlation operation with respect to the reference image and the second reference image of the same image position as the reference image and the first reference image, the correlation performed The second disparity dl (u, v) derived from the correlation calculation result of the unit is output.

  According to this stereo camera unit, the preliminary correlation execution unit and the correlation execution unit can calculate and output the parallax by a single correlation calculation, so that the correlation calculation in stereo matching can be processed at high speed. The preliminary correlation execution unit calculates the parallax by the correlation calculation using the base image and the reference image with a short base length, and since the base length is short, the parallax is small, and each window of the base image and the reference image is set to the same coordinate. Since the parallax can be obtained, the setting of the window is easy. Further, in the correlation execution unit, the base line length is as long as k (k> 1) times, so that the parallax is large, but the base line length is within a range limited by a condition that produces parallax of k times the parallax obtained in the preliminary correlation execution unit Since it is possible to set a long reference image window, it is easy to set the window, and the correlation result is obtained from a long base length set, so that the resolution is increased.

  In the stereo camera unit, the correlation calculation is preferably performed for each pixel.

  The correlation calculation is preferably performed by a correlation calculation method capable of calculating a parallax indicating the maximum correlation by performing a correlation calculation between one window of the standard image and one window of the reference image once. As such a correlation calculation method, there is a phase only correlation method (POC).

Further, in the stereo matching method according to the present embodiment , a reference image is taken by a first camera in which each center position of each image sensor is on the same base line, and a predetermined reference line length is set with respect to the first camera. A first reference image is taken by a second camera provided at a position having L, and a third reference provided at a position having a predetermined reference line length k × L (k> 1) with respect to the first camera. a step of capturing a second reference picture by the camera, pre-set Kimoto quasi image and a window to the first reference image, respectively, the first by a correlation operation with both window run by the phase-only correlation method step a, the window of the first parallax ds (u, v) and k multiplied by k × ds (u, v) the second reference image based on the parallax calculating the parallax ds (u, v) Set the reference image and the first image Wherein executing a correlation operation by the phase-only correlation method with respect to the same image position as the reference image before Kimoto quasi image and the second reference image, a second disparity dl derived from the correlation calculation result (U, v) is output.

  According to this stereo matching method, since the parallax can be output by one correlation calculation for each of the short base length set, the long set of base images, and the reference image, the correlation calculation in stereo matching can be processed at high speed. . Parallax calculation is performed by correlation calculation using a base image and a reference image with a short base length, and since the base length is short, the parallax is small, and the parallax can be obtained even if each window of the base image and the reference image is set to the same coordinate. This makes it easy to set up the window. In addition, in the correlation calculation using the base image and the reference image having a long base length, the parallax is large because the base length is as long as k (k> 1) times. However, the parallax is k times the parallax obtained with the short base length pair. Since it is possible to set a reference image window with a long baseline length within the range limited by the conditions that cause the error, the setting of the window becomes easy and the correlation result is obtained with the long baseline length set, resulting in high resolution. Become.

  In the stereo matching method, the correlation calculation is preferably performed for each pixel.

  The correlation calculation is preferably performed by a correlation calculation method capable of calculating a parallax indicating the maximum correlation by performing a correlation calculation between one window of the standard image and one window of the reference image once. As such a correlation calculation method, there is a phase only correlation method (POC).

  According to the stereo camera unit and stereo matching method of the present invention, the correlation calculation in stereo matching can be processed at high speed.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  <First Embodiment>

  FIG. 1 is a block diagram showing a schematic configuration of a stereo camera unit according to the first embodiment. FIG. 2 is a plan view showing the schematic configuration (a) of the first stereo camera, the schematic configuration (b) of the second stereo camera, and the relative positional relationship between them in the stereo camera unit of FIG.

  As shown in FIG. 1, the stereo camera unit 10 includes a first stereo camera 1 that captures an object and obtains a base image and a reference image having a short base length, and a base image and a reference image that have a long base length. A second stereo camera 2 for obtaining

  In addition, the stereo camera unit 10 includes an image input unit 11 for inputting a set of reference images having a short base length and a reference image from the first stereo camera 1, and a set of reference images having a long base length from the second stereo camera 2. And an image input unit 13 for inputting a reference image.

  Further, the stereo camera unit 10 uses a preliminary correlation execution unit 12 that performs a preliminary correlation calculation process using a base image and a reference image having a short baseline length, and uses the calculation result from the preliminary correlation execution unit 12 to calculate the baseline length. A correlation execution unit 14 that performs a correlation calculation process using a long set of standard images and reference images, and an output unit 15 that outputs a parallax calculation result from the correlation execution unit 14 to the outside.

  As shown in FIG. 2A, the first stereo camera 1 has a pair of cameras 1a and 1b, and the center position of the image sensor of the camera 1a and the center position of the image sensor of the camera 1b are on the base line c. The baseline length g, which is the interval between the center positions, is set to a short value, for example, 10 cg. An image captured by the camera 1a is used as a standard image, an image captured by the camera 1b is used as a reference image, and these are used as a standard image and reference image of a set having a short baseline length.

  2B, the second stereo camera 2 includes a pair of cameras 2a and 2b, and the center position of the image sensor of the camera 2a and the center position of the image sensor of the camera 2b are the base line d. The base line length h, which is the distance between the center positions, is set longer than the base line length g and is, for example, 40 cm. An image captured by the camera 2a is set as a standard image, an image captured by the camera 2b is set as a reference image, and these are set as a standard image and a reference image having a long base line length.

  Each camera 1a, 1b, 2a, 2b is provided with an image sensor composed of a CCD or a CMOS sensor. The image sensor is composed of a large number of light receiving elements and has a photoelectric conversion function, and outputs an image signal for each pixel. .

Here, in FIG. 2, the baseline length ratio k is expressed by the following equation (1).
k = h / g (1)
Since baseline length h> baseline length g, k> 1. In the above example, k = 40/10 = 4.

  The base lines c and d of the stereo cameras 1 and 2 in FIG. 2 are parallel, and the center positions of the cameras 1a and 2a for obtaining the reference image are located on the line p orthogonal to the base lines c and d. That is, the starting points of the base line length g of the first stereo camera 1 and the base line length h of the second stereo camera 2 are located on the same line p orthogonal to the base lines c and d.

  Since the baseline length g of the first stereo camera 1 is shorter than the baseline length h of the second stereo camera 2, the obtained parallax is small and the resolution is low. On the other hand, the second stereo camera 2 has a large parallax obtained as compared with the first stereo camera 1 as k (baseline length ratio) times (k> 1), and has high resolution.

  The preliminary correlation execution unit 12 and the correlation execution unit 14 in FIG. 1 will be described with reference to FIG. FIG. 3 shows a standard image (a) obtained by setting the window with the first stereo camera 1 of FIGS. 1 and 2, a reference image (b), a standard obtained by the second stereo camera 2 and set with a window. It is a figure which shows roughly the image (c) and the reference image (d) similarly.

  The preliminary correlation execution unit 12 preliminarily calculates the parallax using a set of base images and reference images having a short baseline length obtained by the cameras 1 a and 1 b of the stereo camera 1.

  That is, the preliminary correlation execution unit 12 sets a point O1 on a reference image with a short base length for calculating parallax as shown in FIG. 3A, and an n × n window W1 centering on the set point O1. On the other hand, the same size as the window W1 at the same position (point O2) as the window W1 set as the standard image in FIG. 3A on the reference images having a short base line length as shown in FIG. N × n windows W2 are set, and then correlation calculation processing is performed in both windows W1 and W2. The calculation result is the parallax ds (u, v) in the set with the short baseline length, and is output to the correlation execution unit 14. The calculation result is “parallax = 10” as shown in FIG.

  For example, a POC (Phase Only Correlation) method can be used for the correlation calculation process, and the preliminary correlation execution unit 12 includes an integrated element that can execute the calculation process in hardware. The phase-only correlation method is a known stereo matching method using phase information, and a parallax indicating a maximum correlation can be calculated by performing a correlation operation between one window of a standard image and one window of a reference image once. is there.

  The correlation execution unit 14 calculates the parallax using the base image and the reference image having a long base line length obtained by the cameras 2a and 2b of the stereo camera 2 in FIGS. 3A and 3B. The reference image window is set based on the parallax calculation result obtained by the preliminary correlation execution unit 12.

  That is, in the correlation execution unit 14, n is set at the same position (point O3) as the window W1 set as the reference image of FIG. 3A on the reference image having a long base line length as shown in FIG. Xn window W3 is set, and on the other hand, the parallax ds (u, v) obtained by the preliminary correlation execution unit 12 is multiplied by k, and a reference image of a set having a long baseline length as shown in FIG. Then, an n × n window W4 having the same size as the window W3 is set around the point O4 corresponding to the k-times parallax (4 × 10 = 40 in the above example), and then a predetermined parallax range The correlation calculation processing of both windows W3 and W4 is performed. The parallax dl (u, v) that is the calculation result is output to the output unit 15 and output from the output unit 15 to the outside as the output of the stereo camera unit 10. The calculation result is “parallax = 41” as shown in FIG.

  For example, a POC (Phase Only Correlation) method can be used for the correlation calculation process, and the correlation execution unit 14 includes an integrated element that can execute the calculation process in hardware. The phase-only correlation method is a known stereo matching method using phase information, and a parallax indicating a maximum correlation can be calculated by performing a correlation operation between one window of a standard image and one window of a reference image once. is there.

  As described above, in the preliminary correlation execution unit 12 of FIG. 1, the reference image window W <b> 2 of the set with a short base length in FIG. 3B is because the base length of the first stereo camera 1 is short and the parallax is small. 3A can be set around the same point as the window W1 set in the reference image having a short base length in FIG. For this reason, it is easy to set the window W2 for the reference image.

  Further, the reference image window W4 of the set of reference images having a long base length in FIG. 3D has a long base line length of the second stereo camera 2 and has a large parallax, so that the reference image having a long base length in FIG. 1 cannot be set around the same point as the window W3 set in FIG. 1, and it is difficult to set the point because it is difficult to find the point. However, the correlation execution unit 14 in FIG. Based on this, the reference image window W4 having a long base line length is set, so that it is easy to set the reference image window W4 having a long base line length.

  In addition, the parallax obtained by the stereo cameras 1 and 2 is obtained in units of “1” in units of pixels. However, since the second stereo camera 2 has a large base line and a large parallax, the resolution is improved. The obtained parallax is 41 instead of 4 (baseline length ratio) × 10 (parallax obtained with a short baseline length) = 40, and it can be understood that the resolution is improved.

  Next, the operation (steps S01 to S09) of the stereo camera unit 10 of FIGS. 1 to 3 will be described with reference to the flowchart of FIG.

  When the frame processing is started in the stereo camera unit 10, the point O1 on the reference image on which the parallax is calculated as shown in FIG. Is set (S01). Then, an n × n window W1 is set on the reference image around the set point O1 (S02).

  Next, the window W2 is set at the same position O2 as the window W1 set as the standard image as shown in FIG. 3B with respect to the reference images having the short baseline length obtained by the stereo camera 1 (S03). ).

  Next, the preliminary correlation execution unit 12 in FIG. 1 performs correlation calculation processing on the base image window W1 and the reference image window W2 of the set with a short base length in FIGS. 3A and 3B, and the result is obtained. The obtained parallax ds (u, v) is output to the correlation execution unit 14 (S04).

  Next, as shown in FIG. 3C, a set of reference images with a long baseline length acquired by the stereo camera 2 is set at the same position O3 as a window W3 set as a reference image with a short set of baseline lengths. Is set (S05).

  Further, the parallax ds (u, v) obtained by the preliminary correlation execution unit 12 of FIG. 1 is multiplied by k (baseline length ratio), and the reference images having a long base length obtained by the stereo camera 2 are illustrated in FIG. As in 3 (d), the window W4 is set around the point O4 corresponding to the parallax multiplied by k (S06).

  Next, the correlation execution unit 14 of FIG. 1 performs correlation calculation processing in the base image window W3 and the reference image window W4 of the set having a long base length in FIGS. 3C and 3D, and the obtained disparity dl (u, v) is output from the correlation execution unit 14 (S07), and output from the output unit 15 as the output of the stereo camera unit 10 (S08).

  The pixels for calculating the parallax in the windows W1, W2, W3, and W4 described above are moved in the u direction and the v direction in FIG. (S09).

  According to the stereo camera unit 10 of FIGS. 1 to 4, the correlation calculation of the correspondence relationship between the standard image and the reference image in the stereo matching can be performed at a higher speed than when the pyramid image having the pyramid structure is used.

That is, conventionally, in order to obtain the distance data of the output of Su × Sv at intervals of 2 ^P pixels, for example, when the stereo operation is performed on the pyramid image having the pyramid structure using the phase only correlation method (POC), the calculated parallax range is In the case of 0 to L pixels,
When L <2 × 2 ^(t−1) , the number of pyramid stages t (t ≧ 1) is determined, and the POC correlation calculation number N is, for example, p = 3 (when one is output to 8 pixels). The number of operations N1 in the first stage, the number of operations N2 in the second stage, and the number of operations N3 in the third stage need not be changed, and it is necessary to perform the correlation operation of Su × Sv times. It is necessary to perform the correlation operation one time, and the following equation (2) is obtained.
N = N1 + N2 + N3 +... + Nt = 3 × Su × Sv + (1/4) × Su × Sv + (2)

On the other hand, in the present embodiment, the preliminary correlation execution unit 12 performs a calculation process once for each pixel for a set of base images / reference images having a short base length, and the correlation execution unit 14 sets a set of long base lengths. Since the calculation process is performed once for each pixel for the reference image and the reference image, it is only necessary to perform a total of two calculation processes, and the number M of correlation calculations is expressed by the following equation (3).
M = 2 × Su × Sv (3)

  From the above formulas (2) and (3), it is understood that M << N, and the number of correlation operations M according to the present embodiment is considerably smaller than the number of correlation operations N when using a pyramid image, and the correlation operations can be processed at high speed. .

  <Second Embodiment>

  FIG. 5 is a block diagram showing a schematic configuration of a stereo camera unit according to the second embodiment. FIG. 6 is a plan view showing the relative positional relationship of each camera in the stereo camera unit of FIG.

  As shown in FIG. 5, the stereo camera unit 30 captures a reference image of a camera 1c that captures an object and obtains a reference image, a camera 2c that obtains a reference image having a short baseline length, and a reference image that also has a long baseline length. A camera 3c to be obtained.

  The stereo camera unit 30 also includes an image input unit 21 for inputting a reference image from the camera 1c and a reference image having a short base length from the camera 2c, a base image from the camera 1c, and a base line length from the camera 3c. An image input unit 23 for inputting a long set of reference images.

  Further, the stereo camera unit 30 uses a preliminary correlation execution unit 22 that performs a preliminary correlation calculation process using a reference image and a reference image having a short base line length, and uses the calculation result from the preliminary correlation execution unit 22 to calculate the baseline length. A correlation execution unit 24 that performs a correlation calculation process using a long set of standard images and reference images, and an output unit 25 that outputs a parallax calculation result from the correlation execution unit 24 to the outside.

  As shown in FIG. 6, the cameras 1c, 2c, and 3c each have an image sensor formed of a CCD or a CMOS sensor, and the center positions of the image sensors of the cameras 1c, 2c, and 3c are on the same base line e. They are arranged at predetermined intervals. That is, the camera 2c is arranged at a position of a distance g from the camera 1c, and the camera 3c is arranged at a position of a distance h from the camera 1c.

  The distance g between the center positions of the camera 1c and the camera 2c constitutes the base line length and is set to be short, for example, 10 cm. Similarly, the distance h between the central positions of the camera 1c and the camera 3c constitutes the base line length and is set to be long, for example, 40 cm.

  The camera 1c and the camera 2c constitute a first stereo camera, the image captured by the camera 1c is used as a reference image, the image captured by the camera 2c is used as a reference image, and these are set as a reference image with a short baseline length, a reference An image.

  Further, the camera 1c and the camera 3c constitute a second stereo camera, the image captured by the camera 1c is used as a reference image, the image captured by the camera 3c is used as a reference image, and these are the reference images having a long base line length. Reference image. In FIG. 6, the baseline length ratio k can be defined in the same manner as in the above formula (1).

  According to the set of base images and reference images having a short base length obtained by the cameras 1c and 2c, the base length g is shorter than the base length h, so that the obtained parallax is small and the resolution is low. On the other hand, according to the base image and reference image of the long base length set obtained by the cameras 1c and 3c, the parallax obtained is larger than that of the short base length set and the resolution is high. This relationship is the same as in the case of FIG.

  The preliminary correlation execution unit 22, the correlation execution unit 24, and the output unit 25 in the stereo camera unit 30 of FIG. 5 are configured in the same manner as in FIGS. 1 and 3, but as described above, the baseline is the image captured by the camera 1c. 1 and 3 is different from that shown in FIGS. 1 and 3 in that it is used as a common reference image in a short set and a long base set.

  Next, the operation (steps S11 to S18) of the stereo camera unit 30 of FIGS. 5 and 6 will be described with reference to the flowchart of FIG. 7 and FIG.

  When frame processing is started in the stereo camera unit 30, as shown in FIG. 3A, a point O1 on the reference image obtained by the camera 1c for calculating the parallax is set (S11). Then, an n × n window W1 is set on the reference image around the set point O1 (S12).

  Next, for a set of reference images with a short baseline length acquired by the camera 2c, an n × n window W2 is set at the same position O2 as the window W1 set as the standard image, as shown in FIG. 3B. (S13).

  Next, the preliminary correlation execution unit 22 of FIG. 5 performs correlation calculation processing on the base image window W1 and the reference image window W2 of the set having a short base length in FIGS. 3A and 3B, and the result is obtained. The obtained parallax ds (u, v) is output to the correlation execution unit 24 (S14).

  Next, the parallax ds (u, v) obtained by the preliminary correlation execution unit 22 in FIG. 5 is multiplied by k (baseline length ratio), and a reference image of a set having a long baseline length obtained by the camera 3c is illustrated in FIG. As shown in 3 (d), an n × n window W4 is set around the point O4 corresponding to the parallax multiplied by k (S15).

  Next, the correlation execution unit 24 of FIG. 5 performs correlation calculation processing on the reference image window W1 and the reference image window W4 having a long base length in the above-described steps S11 and S12, and the obtained parallax dl (u , V) is output from the correlation execution unit 24 (S16), and output from the output unit 25 as the output of the stereo camera unit 30 (S17).

  The pixels for calculating the parallax in the windows W1, W2, and W4 described above are moved in the u and v directions in FIG. 3 to execute steps S11 to S17, and the above steps are repeated until the total number of processing points is completed (S18). ).

  According to the stereo camera unit 30 of FIGS. 5 to 7, as in the case of FIGS. 1 to 4, when the pyramid image having the pyramid structure is used for correlation calculation of the correspondence between the standard image and the reference image in stereo matching. Can be processed at a higher speed.

  In addition, since the reference image captured by the camera 1c is used in common for the group having a short baseline length and the group having a long baseline length, it is necessary to set a window for the reference image having a long baseline length as shown in FIG. There is no. Further, since the camera 1c is shared, the number of cameras can be reduced as compared with FIGS.

  As described above, the best mode for carrying out the present invention has been described. However, the present invention is not limited to these, and various modifications are possible within the scope of the technical idea of the present invention. For example, the preliminary correlation execution units 12 and 22 and the correlation execution units 14 and 24 shown in FIGS. 1 and 5 may use Fourier transform, discrete cosine transform (DCT) or wavelet transform instead of the POC method. . Further, the correlation calculation in the preliminary correlation execution units 12 and 22 and the correlation execution units 14 and 24 may be executed by software by a CPU (Central Processing Unit).

  1 and 2, a stereo camera having a different base length may be provided, and in FIGS. 5 and 6, a camera is further provided at a different position on the base line e (including its extension line) in FIG. Further, another stereo camera may be configured.

  In this specification, the base line is a straight line connecting two points having accurate coordinates, and the base line length is specifically the distance g between the cameras of the stereo camera as shown in FIGS. , H, which are used for distance calculation by triangulation.

It is a block diagram which shows schematic structure of the stereo camera unit by 1st Embodiment. It is a top view which shows schematic structure (a) of the 1st stereo camera in the stereo camera unit of FIG. 1, schematic structure (b) of the 2nd stereo camera, and relative positional relationship of both. A standard image (a) obtained by the first stereo camera 1 shown in FIGS. 1 and 2 and set with a window, a reference image (b), and a reference image obtained by the second stereo camera 2 and set with a window (c). FIG. 6 is a diagram schematically showing a reference image (d). It is a flowchart for demonstrating operation | movement (step S01-S09) of the stereo camera unit 10 of FIGS. 1-3. It is a block diagram which shows schematic structure of the stereo camera unit by 2nd Embodiment. It is a top view which shows the relative positional relationship of each camera in the stereo camera unit of FIG. It is a flowchart for demonstrating operation | movement (step S11-S18) of the stereo camera unit 30 of FIG. 5, FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1, 2 Stereo camera 1a, 1b, 2a, 2b Camera 1c, 2c, 3c Camera 10 Stereo camera unit 12 Preliminary correlation execution part 14 Correlation execution part 30 Stereo camera unit 22 Preliminary correlation execution part 24 Correlation execution part W1, W2, W3 , W4 Window c, d, e Base line g, h Base line length, interval

Claims (4)

A first camera that captures a reference image in which each center position of each image sensor is on the same baseline, and a first reference provided at a position having a predetermined reference line length L with respect to the first camera A second camera for capturing an image; a third camera for capturing a second reference image provided at a position having a predetermined reference line length k × L (k> 1) with respect to the first camera;
Set before Kimoto quasi image and a window to the first reference image, respectively, the preliminary correlation for calculating a first parallax ds by a correlation operation with both window run by the phase-only correlation method (u, v) The execution part;
Set the window of the first parallax ds (u, v) and k multiplied by k × ds (u, v) the second reference image based on the parallax from the preliminary correlation execution unit, the reference image and and a correlation execution unit for executing the phase-only correlation method a correlation operation with respect to the reference image and the second reference image of the same image position as the first reference image,
A stereo camera unit that outputs a second parallax dl (u, v) derived from a correlation calculation result of the correlation execution unit.   The stereo camera unit according to claim 1, wherein the correlation calculation is performed for each pixel. A second camera provided at a position having a predetermined reference line length L with respect to the first camera, in which each center position of each image sensor is on the same base line and a reference image is taken by the first camera. The first reference image is taken by the step, and the second reference image is taken by a third camera provided at a position having a predetermined reference line length k × L (k> 1) with respect to the first camera. When,
A step of setting prior Kimoto quasi image and a window to the first reference image, respectively, to calculate the first parallax ds by a correlation operation with both window run by the phase-only correlation method (u, v) ,
A window of the second reference image is set based on a parallax of k × ds (u, v) obtained by multiplying the first parallax ds (u, v) by k, and the reference image and the first reference image are set . comprises the steps executed by the phase-only correlation method the correlation calculation for the previous Kimoto quasi image and the second reference image of the same image position as,
A stereo matching method characterized by outputting a second parallax dl (u, v) derived from the correlation calculation result. The stereo matching method according to claim 3 , wherein the correlation calculation is performed for each pixel.

Images