TWI465105B - Automatic alignment method for three-dimensional image, apparatus and computer-readable recording medium - Google Patents

Description

Stereo image automatic calibration method and device and computer readable recording medium

The invention relates to a stereo image automatic calibration method and device and a computer readable recording medium thereof, in particular to an automatic calibration method and device for calibrating left and right images to present a stereo image position error.

The three-dimensional image mainly simulates the parallax seen by the left and right eyes of human beings to produce images with depth of field. Such stereoscopic images can be displayed through a stereoscopic display. The principle is based on the design of the two-eye parallax in the human brain visual processing department. Reconstruction of images produces a three-dimensional sense of visual imagery.

When a stereoscopic image is created, the same object or scene can be captured by two image sensors having a difference in distance between the left and right, and then a single stereo image is combined by image processing, wherein the problem to be solved is between the left and right images. Calibration problem.

Two image sensors (such as cameras) take two images at different angles, generally perform image processing, and perform two steps of calibration at the same time, including brightness, color, and white. Parameters such as white balance, including calibrations on the system, environmental parameters and aberrations, may be part of the calibration procedure.

In the implementation of stereo image calibration, the conventional technology generally captures the image frames of each page through a computer system, and performs image processing one by one, and each page can be temporarily stored in the frame buffer (frame buffer). ), after the left and right images are compared, the correction is performed.

In view of the problem that the conventional technology consumes hardware resources in stereo image processing, the present disclosure proposes a stereo image automatic calibration method and device and a computer readable recording medium thereof, and the technology relates to a hard method for performing stereo image automatic calibration. The body circuit, and includes a computer program that performs automatic calibration using a specific device, and a recording medium that records the program, thereby providing an effective and memory-saving stereo image calibration technique. One of the objects of the present invention is to calibrate the distance between images taken at different positions in a stereoscopic image so that it can present a precise stereoscopic image in an appropriate range.

According to one embodiment of the present invention, a method for automatically calibrating a stereoscopic image includes first capturing a specific object through two directions, and obtaining a left image and a right image through the image sensor, and then selecting an individual region of interest of the left image and the right image ( ROI), forming a left image area and a right image area corresponding to and having a displacement amount.

The method then refers to the hardware limitation, such as the size of the line register, and obtains one-dimensional projection value of the left image area and the right image area pixel value, that is, the characteristic values of the left image area and the right image area respectively, and then stored to the line. Register.

Then calculating a plurality of feature correlation values of the left image region and the right image region, including calculating an average value of the feature values of the left image region and the right image region, calculating a difference between the feature value and the average value of the left image region, and the right image region The difference between the eigenvalue and the average value, and then the difference between the eigenvalue and the average value of the left image region and the right image region is subtracted, and the summed value is added, and the total value is the specific displacement amount at different A plurality of feature association values.

Finally, the minimum feature correlation value of the plurality of feature correlation values corresponding to different displacement amounts is obtained by comparison, and the positions of the left image and the right image are calibrated according to the parallax distance represented by the minimum feature correlation value. Thereby, a better stereoscopic image can be presented.

The above steps may further include unifying the pixel values of the three primary colors in the left image region and the right image region of the obtained region of interest into a pixel value of a single color channel, preferably a value of a green channel, wherein a bilinear inner reference is used. The interpolation method estimates that the red channel corresponds to the pixel value of the green channel, and the blue channel corresponds to the pixel value of the green channel.

In order to calculate the efficiency of the operation, and consider the limitations and costs of the hardware, you can use the technique of binning to effectively reduce the number of pixels actually input by the image of the larger pixel to maintain the specific accuracy. Image processing is performed under the request.

According to one of the embodiments of the present invention, the related stereoscopic image automatic calibration device includes a memory unit for temporarily storing the left image and the right image obtained by the stereo image automatic calibration device, and a color filter unit for separating the image from the red channel, the green channel, and the blue channel. And a numerical operation unit that performs numerical calculation of the pixel value of the image.

The numerical operation unit has a function of integrating the pixel values of the different color channels of the acquired image into an interpolation calculation module of a specific color channel, a combined calculation module for reducing the obtained image pixels, and projecting the image data to the image data. A one-dimensional projection calculation module for obtaining image feature values, and a feature correlation calculation module, the feature correlation calculation module respectively calculates the feature average values of the left image and the right image, and then obtains individual feature values. The difference between the mean values of the respective features and the sum of the features of the left and right images are obtained.

The device specifically includes a memory associated with the memory of the line temporary storage unit and the correlation value obtained by the comparison, obtaining a minimum feature correlation value under different displacements, and performing left image according to the parallax distance represented by the minimum feature correlation value. Correction unit for calibration with the right image.

The embodiment of the present invention further relates to a computer readable recording medium in which a computer program for performing the above-described stereo image automatic calibration method is described.

The stereoscopic image automatic calibration method and device and the computer readable recording medium thereof disclosed in the disclosure relate to a hardware circuit for performing a stereo image automatic calibration method, which uses a specific device to perform an automatic calibration of a computer program, and also relates to a program Record media. One of the objects of the invention is to calibrate the distance between images taken at different locations in a stereoscopic image so that it can present a precise stereoscopic image in an appropriate range.

Figure 1 depicts a schematic diagram of a facility for photographing the same object using two different angles, wherein two left and right image capturing devices 101, 102 (such as a camera or a camera) are illustrated, respectively photographing the same object 103 to obtain two images as shown. (Left image L, right image R), the captured content can be a still image or a dynamic movie that can be processed frame by frame.

2(a), (b) and (c) are next schematic diagrams for obtaining the left and right image characteristic curves. By using a specific device (such as a computer system) to obtain left and right images, as shown in FIG. 2(a), the same object or/and the image data of the left image L and the right image R captured at the same time can be obtained by a processing program. After the two images (L, R) are obtained by the device for performing automatic calibration of the stereo image, the two images can be calculated to have a certain difference, especially the horizontal displacement. To obtain the displacement, a reference line can be set first. The amount of image displacement between the two images is obtained.

According to the embodiment of the method for automatically calibrating a stereoscopic image according to the disclosure, as shown in FIG. 2(b), a region of interest (ROI) having a certain area is obtained in each of the left image L and the right image R. ) are labeled as left image area 201 and right image area 202, respectively.

In the image processing, the pixel values of each region of interest are firstly calculated by one-dimensional projection, and the pixel values are added column by column (in this case, the vertical direction, the same reason can be applied to the lateral accumulation), and a feature value is obtained. (signature), after the calculation of the accumulated value of the entire image area, a distribution curve as shown in FIG. 2(c) may be presented, that is, a schematic diagram for obtaining the feature value by using the region of interest. The present invention estimates the parallax distance between two images based on the feature distribution.

The stereo image auto-calibration method calculates the parallax distance between the left and right images, and calibrates the two images accordingly to obtain a precise stereoscopic image that meets certain requirements, as shown in the flow of FIG. 3, which then describes the stereoscopic image automatic of the present invention. The steps of the calibration method.

Beginning with step S301, the left and right image data are acquired by the image sensor. In the embodiment, two or more cameras at a certain distance can simultaneously capture one target object, and the same object is photographed at different positions by using one camera. The left and right image data is temporarily stored in the memory.

According to the embodiment, in the state that the hardware performance can cope, the entire acquired image can directly perform the subsequent calculation and calibration steps, but the ROI can be separately selected in the left and right images by using the software means as described in step S303. ), each area of interest can be described by coordinates such as x, y, and can be adjusted as needed.

According to the embodiment, the image color is not considered in the need of performing the stereo image position calibration. Therefore, after the image data of the region of interest of the specific coordinate range is obtained in step S305, the image sensor is obtained from the memory. The image color channel in each region of interest sensed, and the embodiment can obtain the three primary colors (R, G, B) pixel values of each pixel through a color filter using a Bayer pattern, which can be referred to. Figure 4 is a schematic view of a color filter used in the present invention. According to one of the embodiments, if the acquired image is not a color image, the step of obtaining the pixel value through the color filter may be ignored.

After obtaining the three primary color pixel values of the color left and right images, in step S307, the color channel values can be replaced by the average value obtained by the surrounding values, thereby unifying the color channels in order to facilitate the operation. For example, referring to FIG. 4 at the same time, the image pixels are composed of color channels such as R (red), G (green), and B (blue), but one of the color channels can be uniformly used for subsequent operations. The correction is based on the G (green) channel.

According to one of the embodiments, in order to facilitate calculation of a block signature for calculating a parallax distance from raw data of an image, the calibration method proposed in the present disclosure adopts a G (green) channel value in a pixel. And use the interpolation method to replace the R (red) and B (blue) channel values with the G channel value, such as a bilinear interpolation. The bilinear interpolation method uses a Bayer CFA pattern (CFA: Color Filter Array) as the basis of the interpolation calculation, that is, the bilinear interpolation method is used to estimate the red channel corresponding to the pixel. The pixel value of the green channel corresponds to the blue channel as the pixel value of the green channel.

For example, please refer to Figure 4. If you want to replace the R1 value in the graph with the G value, that is, R1(G(R1)), this example uses the pixels G1, G3, G4, and G7 values to estimate R1 by interpolation. Value, the equation is as follows:

G(R1)=(G1+G3+G4+G7)/4 ------------- Equation (2)

Taking the B4 value of the pixel as an example, if the B4 value is to be replaced by the G value, the G value around it is interpolated to estimate the value of B4. The equation is as follows:

G(B4)=(G4+G7+G8+G10)/4 ------------ Equation (3)

After unifying the image color channel of the region of interest, if the image size is too hard to process, or is deliberately designed, the present invention can further utilize a pixel binning technique (step S309). In the face of the input image is larger than the hardware can be processed or designed range, a pixel binning technology can be used to effectively reduce the number of image processing actually performed, saving memory usage. The technique of pixel combination is a data pre-processing program, which can combine the image pixels obtained by the sensor according to the size of the memory, and reduce the larger input image to a smaller image and maintain Close to the accuracy of the original image. This pixel combining technique can be applied to any pixel size input image, and the scope of the combination depends on the actual hardware limitations. This step is optional and can be ignored if the image size is not hardware that cannot be processed.

Referring to hardware limitations (such as line register size), pixel combining reduces the size of a larger image to fit the size of the line register. For example, if there is a 4 pixel image, the 2x2 pixel combination step can be used as a pixel to effectively reduce the hardware requirements of image processing, especially the use of memory. According to an example, if the line register is 640 bits and the image is 1280 pixels, the total number of pixels can be increased by two, and the pixels are reduced by half, and then stored in the line register.

After the step of combining the unified color channel and the pixel, the method will be calculated according to the size of the line buffer provided by the hardware according to the size of the line buffer provided by the hardware (in this case, the vertical direction). The one-dimensional projection value of the image preferably directly refers to the image data of the region of interest.

The correlation operation is as follows. In this example, the line buffer is used to store all or a specific part (such as the middle) of 64 pixels (pixel, 0 to 63) of each pixel of the image as an example, and the one-dimensional projection value is obtained by using equation (1), that is, The feature values of the left image region and the feature value of the right image region in the ROI are respectively reflected, and the characteristic curves of the left and right image regions can be reflected.

The images obtained by the different or the same sensors in the left and right directions are respectively labeled as the left image (L) and the right image (R), and the image can be obtained while the region of interest is used as a follow-up operation. After the preliminary processing described above, the subsequent steps are performed according to the region of interest. The acquisition of the interest area includes first setting a square (for example, a square, a rectangle), that is, defining a region of interest (ROI), wherein the X and Y directions are adjustable, and can be referred to FIG. The two frames are respectively indicated by squares, and each column in the block (such as the sum of vertical pixels) is described by the curve C _x (in this case, the image of the horizontal i and the vertical 64 pixels):

Where x is the horizontal coordinate direction; i index indicates the index of the horizontal pixel of the selected image; j is the vertical pixel index, in this case the sum of 64 pixels in the middle of the image or other parts; P(i, j, L) and P(i, j, R) are the pixel values of the coordinates (i, j) in the left and right frames, respectively. Through the summation operation, the entire planar image projection can be represented by one-dimensional data, that is, the characteristic values of the respective images, and a characteristic curve is formed in terms of distribution.

A schematic diagram of obtaining a one-dimensional projection value as shown in FIG. 5(a)(b) can be simultaneously referred to, and FIG. 5(a) is a schematic diagram showing an accumulation of pixel values in a vertical direction of an image range, and then the accumulated values of the respective columns will be reflected as shown in the figure. On the characteristic curve of 5(b), that is, the variation curve (C _x ) of the eigenvalue calculated by equation (1).

After one-dimensional projection, the image features can be represented by a curve, and the correlation value can also conform to the size of the line buffer. As described in step S313, the calculated feature values are temporarily stored in the line buffer. Using this method of projecting on the online register can greatly reduce the required memory.

With continued reference to step S315 of FIG. 3, the minimum feature correlation value is calculated.

In order to obtain the parallax distance between the left frame (L) and the right frame (R), the method proposed by the present disclosure obtains the feature correlation value by using the feature values of the left and right frames and the average value, and then obtains a minimum feature. The correlation signature is related to the characteristic curve obtained by the above left and right frames respectively, as calculated by the following equation (4):

In this case (but the actual implementation is not limited to this example), the pixel values of all or part of the 64 pixels (0 to 63) in the horizontal (x direction) of the selected image area are also used, and the summed The one-dimensional projection value (eigenvalue) is used to calculate the transverse feature average of the image region (Average _x (L/R)); where q is a displacement between the two image regions, which can be 192 in this example. The range of prime (0 to 191) is a displacement amount indicating a horizontal eigenvalue (such as C _x ) of two images in one region (such as ROI), which may be a shift vector, and the displacement vector is Causes the parallax in the horizontal direction between the left and right images. In this way, the mean values of the left and right frames (Average _x (L) and Average _x (R)) are obtained separately, and then the difference between the horizontal and horizontal feature values is calculated by the equation (C _x ( i, L) - Average _x (L) and C _x (i + q, R) - Average _x (R)).

Then subtract the difference between the feature value of the left image region and the right image region and the mean value of the feature, and then add the subtracted value to the total value, that is, the difference between the feature value of the left and right frame and the mean value of the feature is subtracted. Value, the total value is a plurality of feature correlation values R(q) at different q values (displacement amounts), which is the right image region outside the eigenvalue of the left image region (L) and having a distance (R) The shift between the eigenvalues is the associated value of q. In this case, there are 192.

By comparing the feature correlation values corresponding to different displacement amounts, a minimum feature correlation value can be obtained, represented by R _min (q), and the correlation value calculated by the above is taken out as a minimum value in an image region, that is, original The left and right frames are designed to have a displacement q, but the displacement q produces a parallax distance represented by the minimum feature correlation value in the horizontal direction (step S317), that is, a parallax distance that needs to be compensated when the stereoscopic image is presented. Then, calibration is performed in accordance with the parallax distance represented by the minimum feature correlation value R _min (q) (step S319).

One of the purposes of the method for automatically calibrating a stereoscopic image proposed by the above disclosure is to estimate the parallax distance of two images taken from two sensors, and perform calibration according to the reference. The dimensional projection technique projects a pixel value in a specific range onto a characteristic curve, and the steps can be referred to the steps described in the flow of FIG.

If the image is colored, you can input the image of the unified channel first, as in step S601. For example, the red and blue pixels in the R, G, and B color spaces are represented by green values by interpolation; if they are non-colored Image, or you can ignore this step.

Then, according to actual needs, a specific region of interest in the left and right images may be determined (step S603), and the pixel values are accumulated row by row for the pixel values in the specific range (step S605), and a subsequent need for calculating the feature association is generated. The dimension projection value (step S607).

The flow shown in Figure 7 describes the steps of the present invention for calculating feature associated values. Step S701 describes using the calculated one-dimensional projection value of the left and right image interest regions, and then calculating the average value of the lateral features of the left and right image interest regions according to the example shown in the above equation (4), which can be respectively defined as the left image. The first average value and the second average value of the right image (Average _x (L/R), step S703) are subtracted from the horizontal feature values in the left and right images, and the one-dimensional projection values and their relative average values in the left and right images of each line are obtained. The difference may be defined as a second difference between the first difference of each projection value of the left image and each projection value of the right image (step S705).

Then, in step S707, the difference amount (first and second difference) between each lateral feature value and the feature average value is accumulated, and thus the feature correlation value on different displacement vectors (q values) in each image lateral position is obtained (step S709), after the repeated calculation, obtain the minimum feature correlation value (step S711), the value is regarded as the parallax distance, that is, the value to be calibrated, and the image sensor or related device for the left and right image shooting should be calibrated for this need The value adjusts the device distance.

Fig. 8 is a block diagram showing the function of the stereoscopic image automatic calibration apparatus for realizing the above calibration method.

The device itself includes a memory unit 803, a color filter unit 805, a numerical operation unit 807, a line temporary storage unit 809, a comparison unit 811 and a correction unit 813, and the numerical operation unit 807 includes a pixel that can be implemented by software or hardware. The combined computing module 821, the interpolation computing module 823, the one-dimensional projection computing module 825 and the feature correlation computing module 827 further include an area of interest coupled to the memory unit 803 for using data of a specific portion of the image. Module 831.

According to the embodiment, the images in the left and right directions of the specific object can be simultaneously acquired by the image sensors at different positions, and if the static objects are photographed, the same image sensor can be photographed at different positions. As shown in the figure, the stereo image auto-calibration device obtains left and right images by the image sensor 801 and temporarily stores them in the memory unit 803 of the device. The temporarily stored images can be separated from the pixels by the color filter unit 805. ), green channel (G) and blue channel (B). When the image data is acquired, the region of interest module 831 implemented by the software or the hardware can acquire the image of interest (ROI) having a certain area in the left and right images R by using the image stored in the memory unit 803. Calculate distance errors and save computing resources.

Since the calibration method proposed in the present disclosure does not utilize the color characteristic, the pixels can be unified on the same color channel, for example, the above-mentioned interpolating method is used to unify each color channel into a green channel. The pixel values processed by the color filter unit 805 are processed by the numerical operation unit 807.

Then, the numerical operation unit 807 performs numerical operation on the pixel values of the acquired image, and the functions required according to the present invention are mainly divided into a plurality of modules implemented by software or hardware, including a picture in which the acquired images are combined by pixels. The merging calculation module 821, the interpolation calculation module 823 for calculating a single color channel of the image, the one-dimensional projection calculation module 825 for accumulating the image pixels to obtain the one-dimensional eigenvalue, and the feature for calculating the correlation value of the feature according to the features of the left and right images Correlation calculation module 827.

The interpolation calculation module 823 is used to unify the pixel values of the different color channels of the obtained image into a specific color channel, and the above is preferably converted into a green channel, but the actual operation does not exclude unifying to other color channels according to requirements. For the calculation method, refer to equation (2) (3) above.

Then, because of limitation of hardware resources (especially memory), or in order to save hardware resources, the obtained image data may be binned by binning using the pixel combination calculation module 821 in the numerical operation unit 807. The steps effectively reduce the pixels that actually perform the image processing, that is, reduce the larger input image to a smaller image.

When processing the left and right images, the one-dimensional projection calculation module 825 in the numerical operation unit 807 can project the image data of the entire selected portion (such as ROI) onto the data on the one-dimensional coordinates, thereby obtaining a kind of image. The characteristic curve (having a sequence of eigenvalues) is calculated as Equation (1). The device has a temporary storage unit for temporarily storing the processing data of the numerical operation unit 807, such as the line temporary storage unit 809 shown in the figure, and the projected image data can be stored in the first-line temporary storage unit 809, which is different from the general need. The image is temporarily stored in a specific memory (such as Synchronous Dynamic Random Access Memory (SDRAM) in the device), and the use of the line temporary storage unit can save memory for internal computing.

After the image data on the one-dimensional coordinates is obtained by the numerical operation unit 807, the feature-related calculation module 827 calculates the average value of the horizontal and horizontal image pixels, and obtains the difference between the feature values of the left and right images and the average value (respectively It can be defined as the first difference and the second difference), and the difference is the characteristic correlation value of the two, and can be referred to the calculation of the above equation (4).

Then, the comparison unit 811 compares the correlation values of the features to obtain a minimum feature correlation value of the left and right images, and the minimum feature correlation value is represented as a parallax distance, that is, a value required to calibrate and compensate the left and right images to form a stereoscopic image. The calibration of the left and right image positions is performed by the correction unit 813 to obtain a parallax that can preferably present a stereoscopic image. After the image is adjusted, the image is output to the external device 815.

The device for performing automatic calibration of the stereo image proposed by the disclosure is as described in FIG. 8 above. After the image data is acquired by the image sensor, the color channel data of each pixel is obtained through the color filter, and then the region of interest is Calculate and unify the color channel and calculate the one-dimensional projection value to facilitate temporary storage in the line temporary storage unit, and finally calculate the minimum feature correlation value, and then perform image or device correction based on this.

In summary, the stereo image auto-calibration method and related device proposed by the present disclosure aims to perform automatic calibration by estimating the parallax distance of the two images obtained by the two sensors, in particular through uniform color and pixel. The execution of the merge associated with the minimum feature effectively yields the parallax distance and performs calibration accordingly. The automatic calibration method effectively uses the line register to make error judgment by projecting into a one-dimensional space, which can reduce memory usage.

However, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Therefore, equivalent structural changes that are made by using the specification and the contents of the present invention are equally included in the present invention. Within the scope, it is combined with Chen Ming.

101,102. . . Image capture device

103. . . object

L. . . Left image

R. . . Right image

201. . . Left image area

202. . . Right image area

801. . . Image sensor

803. . . Memory unit

805. . . Filter unit

807. . . Numerical arithmetic unit

809. . . Line temporary storage unit

811. . . Alignment unit

813. . . Correction unit

821. . . Pixel combined computing module

823. . . Interpolation calculation module

825. . . One-dimensional projection computing module

827. . . Feature correlation calculation module

831. . . Interest area module

815. . . External device

Step S301 to step S319 automatic calibration process of stereo image

Step S601 to step S607 One-dimensional projection value calculation flow

Step S701 to step S711 feature correlation value calculation process

Figure 1 depicts a schematic diagram of a facility for photographing the same object using two different angles;

2(a), (b) and (c) are schematic diagrams showing the use of the region of interest to obtain feature values according to the present invention;

The flow shown in FIG. 3 describes the steps of the automatic image calibration method of the present invention;

Figure 4 is a schematic view showing a color filter used in the present invention;

Figure 5 (a) (b) shows a schematic diagram of obtaining a one-dimensional projection value according to the present invention;

The flow shown in FIG. 6 describes the steps of the present invention for calculating a one-dimensional projection value;

The flow shown in Figure 7 describes the steps of calculating the associated value of the feature of the present invention;

FIG. 8 is a functional block diagram of a stereo image automatic calibration apparatus according to the present invention.

S301. . . Get left and right image data

S303. . . Select interest area

S305. . . Get image color channel

S307. . . Unified color channel

S309. . . Pixel merge

S311. . . Get one-dimensional projection values column by column

S313. . . Temporarily stored in the line register

S315. . . Calculate the minimum feature correlation value

S317. . . Left and right parallax distance

S319. . . Perform left and right image calibration

Claims (10)

A method for automatically calibrating a stereoscopic image includes: obtaining an image of a left image and a right image of an object through an image sensor; and selecting an individual region of interest of the left image and the right image to form a correspondence and having a displacement amount of one left An image area and a right image area; referring to the size of the line register, obtaining one of the left image area and the right image area pixel value, that is, the feature value of the left image area and the right image area respectively a feature value; storing one of the left image region and the right image region pixel value to the line register, wherein the one-dimensional projection value is an eigenvalue of the left image region and the right image region; a plurality of feature correlation values of the left image region and the right image region, including: calculating a first average value of the feature values of the left image region, and a second average value of the feature values of the right image region; calculating the left a first difference between the feature value of the image region and the first average value; a second difference between the feature value of the right image region and the second average value; the first difference and the second difference And subtracting the value of the first difference from the second difference, wherein the total value is the plurality of feature correlation values at different displacement amounts; and obtaining the plurality of feature associations corresponding to the different displacement amounts One of the values of the minimum feature correlation value; and calibrating the position of the left image and the right image based on the parallax distance represented by the minimum feature correlation value. The method for automatically calibrating a stereoscopic image according to claim 1, wherein if the left image and the color image of the right image are used, the pixels of the left image region and the right image region of the acquired region of interest are unified. The pixel value for a single color channel. The method for automatically calibrating a stereoscopic image according to claim 2, wherein the left image region and the pixel region of the right image region obtain a three primary color pixel value of each pixel by using a Bayer color filter. The method for automatically calibrating a stereoscopic image according to claim 3, wherein the pixel values of the pixels are unified into the pixel values of the green channel. The method for automatically calibrating a stereoscopic image according to claim 4, wherein the step of the unified color channel estimates a pixel value corresponding to a green channel in a pixel by a bilinear interpolation method, and blue The color channel corresponds to the pixel value of the green channel. The method for automatically calibrating a stereoscopic image according to claim 1, wherein after the step of selecting the left image and the individual region of interest of the right image, the method further comprises: reducing the actual execution of the left by a pixel combining technique The number of pixels in the image area and the right image area. A computer readable recording medium, wherein a computer program is stored, the program performs a stereo image auto-calibration method, the method comprises: obtaining an image of a left image and a right image of an object through an image sensor; and selecting the left image The image and the individual region of interest of the right image are corresponding to each other and have a left image region and a right image region; and the pixels of the left image region and the right image region of the acquired region of interest are unified into a single image. The pixel value of the color channel; the pixel combination method reduces the number of pixels actually executing the left image area and the right image area; referring to the size of the line register, the left image area and the right image area are drawn a one-dimensional projection value of the prime value, that is, a feature value of the left image region and a feature value of the right image region; storing one of the left image region and the right image region pixel value to the line temporary storage The one-dimensional projection value is a feature value of the left image region and the right image region; and calculating a plurality of feature correlation values between the left image region and the right image region The method includes: calculating a first average value of the feature values of the left image region, and a second average value of the feature values of the right image region; and calculating a first difference between the feature value of the left image region and the first average value; And a second difference between the feature value of the right image region and the second average value; subtracting the first difference from the second difference; summing the value of the first difference and the second difference, adding The total value is the plurality of feature correlation values at different displacement amounts; obtaining a minimum feature correlation value corresponding to one of the plurality of feature correlation values different from the displacement amount; and a parallax distance represented by the minimum feature correlation value Calibrate the position of the left image and the right image. A stereoscopic image automatic calibration device comprises: a memory unit, temporarily storing the stereo image automatic calibration device to obtain one of a left image and a right image; and a color filter unit for respectively displaying a red channel, a green channel and a blue image a color channel; a numerical operation unit that performs numerical operations on the pixel values of the image, including: an interpolation calculation module that unifies the pixel values of the different color channels of the obtained image into a specific color channel; a group, the pixel of the obtained image is reduced; the one-dimensional projection calculation module projects the image data onto the one-dimensional coordinate to obtain the feature value of the image; and a feature correlation calculation module calculates the left image and the right image respectively The characteristic mean value, and then obtain the difference between the characteristic value of the left image and the right image and the characteristic mean value of the respective right image, and after summing up, obtain the feature correlation value between the left image and the right image; the first line temporary storage unit, Temporarily storing the data of the numerical operation unit, including the obtained image feature value; and comparing the value of the left image to the feature of the right image, The minimum feature association value; and a correcting means for calibrating the left image and the right image of the parallax from the minimum value represented by the associated feature. The stereoscopic image automatic calibration device of claim 8, wherein the left image and the right image processed by the device acquire an area of interest via an area of interest module upon input. The stereoscopic image automatic calibration device according to claim 8, wherein the line temporary storage unit is one of the stereoscopic image automatic calibration devices and the synchronous dynamic random access memory.