CN112991517B - A 3D reconstruction method for automatic matching of texture image encoding and decoding - Google Patents

Abstract

The invention discloses a three-dimensional reconstruction method based on texture-coded images, comprising: step 1, projecting a color-coded image generated by M-array coding onto an object to be measured, and using a camera to photograph the area to be measured to obtain left and right images. In step 2, the target area is extracted based on the Hough circle detection method and the perspective transformation principle, so as to avoid the influence of environmental debris on the projection area. Step 3: Convert the color information of the image into a code value based on image enhancement and color recognition preprocessing based on color migration technology. In step 4, the matching point pairs are obtained by directly decoding the M-array encoding method. Step 5: Using the obtained point pairs with the same name, perform three-dimensional reconstruction based on the principle of stereo vision, and obtain the three-dimensional coordinates of the spatial points corresponding to the two-dimensional points. The invention combines spatial coding and decoding and binocular vision structured light three-dimensional reconstruction technology to improve the speed and accuracy of three-dimensional reconstruction, and provides a new idea for high-efficiency three-dimensional reconstruction on the premise of maintaining high simulation degree.

Description

A 3D reconstruction method for automatic matching of texture image encoding and decoding

technical field

The invention belongs to the field of image encoding and decoding and structured light three-dimensional reconstruction. The left and right images with enhanced texture are obtained by photographing the object to be tested through the projected encoding pattern, and the left and right image homonymic points are directly located through a decoding algorithm, thereby realizing fast and high-precision three-dimensional reconstruction.

Background technique

With the gradual development of science and technology, computer vision technology has also continued to improve. Among them, 3D reconstruction technology has developed rapidly and gradually improved, and has been widely and effectively applied in various fields. However, the traditional 3D reconstruction method has many limitations when applied in production practice. How to ensure the robustness of the matching in the 3D reconstruction process, so as to make the reconstruction high reliability and improve the efficiency of the reconstruction process, is currently being studied by scholars in related fields. focus. The three-dimensional reconstruction technology based on encoded structured light uses a projector to map the encoded texture pattern to the surface of the object to be measured, and then uses the camera as a sensor to capture the image of the object to be measured. The three-dimensional depth information is geometrically deformed. Using digital image processing technology to decode the image can directly locate the points to obtain matching point pairs. The encoding and decoding process in this method replaces the error-prone, complex and time-consuming feature extraction and In the process of feature matching, the speed of the three-dimensional reconstruction process is fast, and the obtained three-dimensional reconstruction results have high precision. Obviously, the encoding and decoding algorithm is the key to this method, which determines the efficiency of 3D reconstruction. Moreover, in recent years, various fields have higher requirements for 3D reconstruction technology. It is hoped that the 3D reconstruction of moving objects can be achieved while maintaining high precision. Reconstruction, which can only use a single image for projection enhancement, so the encoding and decoding algorithm for a single image has become an inevitable trend of research, and it is also the focus and difficulty of research and application.

SUMMARY OF THE INVENTION

Aiming at using a single M-array coded image for structured light three-dimensional reconstruction, the present invention proposes an extraction target area based on perspective transformation, image preprocessing based on color migration, decoding algorithm based on M-array coding, and three-dimensional reconstruction based on stereo vision. The method realizes the combination of spatial encoding and decoding technology and stereo vision 3D reconstruction technology, which avoids the difficulty and error-prone point of feature point matching in traditional 3D reconstruction methods, and improves the speed of the 3D reconstruction process and the accuracy of the 3D reconstruction results. It provides new ideas for efficient 3D reconstruction under the premise of maintaining high fidelity.

In order to achieve the above purpose, a three-dimensional reconstruction method based on texture image coding designed by the present invention comprises the following steps:

Step 1), project the color-coded image generated by the M-array coding matrix onto the object to be measured, and use a camera to photograph the area to be measured to obtain a left image and a right image.

Step 2), extracting the target area of the image and the right image projection based on the Hough circle detection method and the perspective transformation principle, so as to avoid the influence of environmental debris on the projection area;

Step 3), based on the color migration technology, the left image and the right image after the extraction target area are respectively enhanced and color recognition is carried out, the color information of the image is converted into a code value, and the left image color recognition diagram and the right image color recognition diagram are obtained;

Step 4), for the method for M array coding, the color recognition diagram obtained in step 3) is directly decoded to obtain a pair of matching points;

Step 5), using the obtained matching point pairs to perform three-dimensional reconstruction based on the principle of stereo vision, to obtain the three-dimensional coordinates of the spatial points corresponding to the two-dimensional points.

Further, the specific implementation mode of step 1) is as follows;

First, the designed single color-coded image is projected onto the surface of the object to be tested through a projector, and then the camera is used to take images of the object to be tested from different angles to obtain a left image and a right image. The color-coded image is encoded by an M-array. Matrix generation.

Further, the specific implementation mode of step 2) is as follows;

First, the gradient edge detection is performed on the image to obtain the binary image of edge detection; secondly, the Hough transform is used to detect the circle function, and the appropriate three input parameters of minimum distance, maximum radius and minimum radius are selected, and the camera image is obtained by detection. The coordinates of the center of the four circles, that is, the coordinates of the four corners of the projection area; finally, the detected coordinates of the four corners are taken as the original coordinates, and the coordinates of the four corners of the designed color-coded image are taken as the coordinates of the four corners after the perspective transformation. The coordinates of these four groups of two-dimensional mapping points are substituted into the perspective transformation equation system, and the values of the eight unknowns are calculated to obtain the perspective transformation matrix of the target area of the left and right image extraction.

Further, the specific implementation mode of step 3) is as follows;

First, the color-coded image is used as the target image for color migration, and the image after extracting the target area is used as the original image for color migration. The target image and the original image are converted from RGB color space to lαβ color space, and the approximately orthogonal luminance component and two chrominance components α, β;

Next, calculate the mean and standard deviation of the two images in the lαβ space respectively, first subtract the mean value of the lαβ channel of the target image from the data of the lαβ channel of the target image, and then scale the obtained new data proportionally, the scaling factor is the original The ratio of the standard deviation of the image to the standard deviation of the target image, and finally adding the obtained results to the mean value of the original image's lαβ channel, that is, to obtain the original image after color migration processing with the same mean and variance as the target image;

Finally, color recognition is performed on the image after color migration; the image after color migration is converted from the lαβ color space back to the RGB color space, the RGB value of each pixel is extracted, and the pixel is classified by the discriminant sentence. If the gray value is greater than the threshold t1, it is recognized as white. If the gray value of the red channel is the largest and the gray value difference with the other two channels is greater than t2, it is recognized as red, green and blue. The recognition principle is the same; for the pixels of the three colors recognized, they are represented by the corresponding "0", "1", and "2" when designing the coding pattern, and the white color is represented by "3", and an image-sized matrix is obtained. , that is, the color identification map is obtained.

Further, the specific implementation mode of step 4) is as follows;

According to the color recognition map of the left image, a right neighborhood map and a lower neighborhood map are respectively constructed; the process of constructing the right neighborhood map is to retrieve each pixel to the right in the color recognition map until the white border is retrieved. For the first primitive pixel, record the code value and coordinates of the point in the right neighborhood graph. The process of constructing the lower neighborhood graph is to search down for each pixel in the color recognition graph until the white border is retrieved. After the first primitive pixel, record the code value and coordinates of the point in the lower neighborhood map; the above-mentioned method is based on the situation that the pixel point is located inside a single color block, if the pixel point is located on the white border, then Retrieve to the right or down along the point until the pixel point with the first code value "0" or "1" or "2" is retrieved, and this point is used as the starting point;

Secondly, for the to-be-determined points in the left image, use the color identification map, the right neighborhood map and the lower neighborhood map to find the right primitive that is the most adjacent to the current primitive in the right neighborhood map. Find the primitive most adjacent to the current primitive, construct a 3×3 window where the current primitive is located, and put this window into the M-array coding matrix corresponding to the color-coded image to perform template matching to realize window positioning and primitives. Positioning, the primitive refers to a single color block in the color-coded image, the value of the encoding matrix corresponds to each primitive of the color-coded image, and "0", "1", and "2" are respectively associated with the design code The primitives of red, green and blue in the image correspond to each other;

Once again, the interior of the primitive is further accurately positioned through the positional relationship between the pixel coordinates of the to-be-located point and the white border around the primitive. If the to-be-located point is located inside the primitive, the pixels to be located are retrieved in four directions, up, down, left, and right, until the white border is encountered. The frame stops, that is, when the first point with the code value of "3" is retrieved, it stops, calculates the number of pixels that pass through the retrieval in the four directions, and realizes positioning through simple calculation; if the point to be determined is not inside the primitive, the Search right or down until the first point inside the primitive is retrieved within the threshold range, and use this point as the starting point, and then perform precise positioning;

Finally, the matching point of the to-be-located point is searched in the right image, which is similar to the steps of primitive positioning and intra-unit accurate positioning. First, the primitive is located according to the position of the to-be-located point in the left image, and then further intra-unit accurate positioning is performed; Among them, the geometric constraints of perspective transformation are used to determine the starting position of the search when the primitive is located, and the search is only performed within a certain range. If no matching primitive is found within the range, the matching of the to-be-located point is discarded.

Further, the specific implementation mode of step 5) is as follows;

First, use the MATLAB camera calibration toolbox to calibrate the camera internal parameter matrix, make a checkerboard calibration picture as the calibration board, use the camera to shoot the calibration board from different angles, get as many pictures as possible, and measure the checkerboard on the calibration board with a ruler. The actual side length of the small square, the unit is millimeters, input the obtained data into the calibration toolbox to get the calibration result of the camera's internal parameters;

Secondly, the eight-point algorithm of RANSAC is used to calculate the value of the fundamental matrix, and eight point pairs are randomly selected from the point pairs of the same name, and the result of the fundamental matrix is calculated by solving the linear equation system, and then the support is found in all the original point pairs. For the point pairs of the calculated fundamental matrix, if the number of supported point pairs is large enough, the result of the calculated fundamental matrix is considered to be credible, and all the supporting points are pinched by the least squares method to obtain the final fundamental matrix As a result, on the contrary, if only a small number of matching point pairs satisfy the basic matrix results obtained by the initial calculation, repeat the above steps until the optimal solution is found; after obtaining the optimal solution of the basic matrix, the basic matrix and the camera internal parameter matrix are calculated to obtain the result of the essential matrix;

Finally, the essential matrix is only related to the relative position and attitude of the camera when two pictures are taken. The essential matrix is decomposed by SVD to obtain the transformation matrix [R|t] representing the right image relative to the left image, and four results are obtained from the calculation, There is only one of these four results, so that the points calculated by the triangulation principle all fall in front of the two cameras, and the only result is obtained by taking the point calculation and verifying.

The present invention has following positive effect:

1) The present invention proposes a decoding method based on a single M-array encoding scheme, by separately obtaining the coordinates of the points in the left and right images in the encoding pattern, the coordinates are the same, that is, matching point pairs, and this process replaces the traditional three-dimensional reconstruction. The algorithm of feature point matching, which is complex, time-consuming and prone to mis-matching in the process, improves the accuracy and efficiency of reconstruction, and only requires one coded image to achieve positioning, and it is also suitable for moving scenes.

2) Before decoding the encoded image, the present invention first uses perspective transformation to extract the target, which avoids the influence of environmental debris on the projection area, and facilitates the matching of the left and right images with the same name after subsequent decoding.

3) The present invention proposes a method for preprocessing using color migration technology, which solves the problem that the object is affected by its own surface texture and ambient lighting, resulting in low decoding accuracy. The judgment can realize the color recognition.

The invention can realize the three-dimensional reconstruction of the object to be measured more accurately and stably, combines the spatial coding and decoding technology and the binocular vision structured light three-dimensional reconstruction technology, and integrates the advantages of the two to improve the speed of the three-dimensional reconstruction process and the accuracy of the three-dimensional reconstruction results. precision.

Description of drawings

Figure 1 is a color-coded image for projection in the present invention.

Figure 2 is a flow chart of the present invention.

FIG. 3 is a flow chart of making a neighborhood graph in the present invention.

FIG. 4 is a flow chart of performing window template matching to realize primitive positioning in the present invention.

FIG. 5 is a schematic diagram of precise positioning within a primitive in the present invention.

Detailed ways

In order to make the objectives, technical solutions and effects of the present invention clearer and clearer, the present invention will be described in further detail below with reference to the accompanying drawings.

Embodiment 1, a three-dimensional reconstruction method based on texture image coding, comprising the following steps:

Step 1), in this step, a texture-encoded image of the object to be tested that has been enhanced by projection is captured for subsequent processing. First, the designed single color-coded image is projected onto the surface of the object to be tested through a projector, and then the camera is used to take images of the object to be tested from different angles to obtain a left image and a right image. The color-coded image is encoded by an M-array. Matrix generation.

Step 2), using the perspective transformation principle to extract the target area of the projection of the left image and the right image, so as to avoid the influence of environmental debris on the projection area.

First, the gradient edge detection is performed on the image to obtain the binary image of edge detection.

Secondly, use the Hough transform packaged in OpenCV to detect the circle function, select the appropriate three input parameters of minimum distance, maximum radius, and minimum radius, and detect the coordinates of the center of the four circles in the image captured by the camera, that is, the projection. The coordinates of the four corners of the area.

Finally, take the detected coordinates of the four corners as the original coordinates, take the coordinates of the four corners of the designed color-coded image as the coordinates after perspective transformation, and substitute the coordinates of these four sets of two-dimensional mapping points into the perspective transformation equations, Eight unknowns are obtained by calculation, and the perspective transformation matrix of the target area of the left and right image extraction is obtained.

Step 3), using the color migration technology to enhance the left image and the right image after the extraction of the target area and perform color recognition, and convert the color information of the image into a code value to obtain a color recognition map.

First, take the color-coded image as the target image for color migration, take the image after extracting the target area as the original image for color migration, and convert the target image and the original image from the RGB color space to the lαβ color space (approximately orthogonal The luminance component l and two chrominance components α, β).

Next, calculate the mean and standard deviation of the two images in the lαβ space respectively, first subtract the mean value of the lαβ channel of the target image from the data of the lαβ channel of the target image, and then scale the obtained new data proportionally, the scaling factor is the original The ratio of the standard deviation of the image to the standard deviation of the target image, and finally the obtained results are added to the mean value of the lαβ channel of the original image, that is, the original image after color migration processing with the same mean and variance as the target image is obtained.

Finally, color recognition is performed on the image after color migration processing. Convert the image after color migration from the lαβ color space back to the RGB color space, extract the RGB value of each pixel, and classify the pixels through simple discriminant sentences. In this example, if the grayscale values of the three channels are all greater than 50, It is recognized as white. If the gray value of the red channel is the largest and the gray value difference between the red channel and the other two channels is greater than 60, it is recognized as red. The recognition of green and blue is the same as that of red. For the pixels of the three colors identified, they are represented by "0", "1", and "2" corresponding to the design of the coding pattern, and the white color is represented by "3", and an image-sized matrix is obtained, that is, the color is obtained. Identify the graph.

Step 4), directly decoding the color identification map obtained in step 3) according to the method of M array encoding to obtain matching point pairs.

First, as shown in Figure 3, according to the color recognition map of the left image, a right neighborhood map and a lower neighborhood map are respectively constructed. The process of constructing the right neighborhood map is to search to the right for each pixel in the color recognition map until the first primitive pixel after the white border (code value "3") is retrieved, in the right neighborhood map Record the code value and coordinates of the point, and the process of constructing the neighborhood map is to search down for each pixel in the color recognition map until the first one after the white border (code value "3") is retrieved. The primitive pixel, the code value and coordinates of the point are recorded in the lower neighborhood graph. The above-mentioned method is based on the situation that the pixel is located inside a single color block. If the pixel is located on the white border, search to the right or down along the point until the first code value "0" or "1" is obtained. ” or “2” pixels, use this point as the starting point.

Secondly, as shown in Figure 4, for the to-be-determined point in the left image, using the color identification map, the right neighborhood map and the lower neighborhood map, by finding the right primitive that is the most adjacent to the current primitive in the right neighborhood graph, in the lower Find the most adjacent primitive below the current primitive in the neighborhood graph, construct a 3×3 window where the current primitive is located, and put this window into the M-array coding matrix corresponding to the color-coded image to perform template matching to realize the window. Positioning and primitive positioning, the primitive refers to a single color block in the color-coded image, the value of the encoding matrix corresponds to each primitive of the color-coded image, "0", "1", "2" They correspond to the primitives of the three colors of red, green, and blue when designing an encoded image.

Again, as shown in Figure 5, the interior of the primitive is further accurately positioned by the positional relationship between the pixel coordinates of the to-be-determined point and the white border around the primitive. If the to-be-located point is located inside the primitive, the pixels to be located are searched in four directions, up, down, left, and right, until it stops when the white border is encountered, that is, the point with the first code value of "3" is retrieved, and the four directions are calculated. The number of pixels passed through respectively during retrieval, and positioning is achieved through simple calculation. If the to-be-determined point is not inside the primitive, search to the right or down until the first point inside the primitive is retrieved within the threshold range, and use this point as the starting point, and then perform precise positioning.

Finally, the matching point of the to-be-located point is searched in the right image, which is similar to the steps of primitive positioning and intra-unit precise positioning. First, the primitive is located according to the position of the to-be-located point in the left image, and then further intra-unit accurate positioning is performed. Among them, the geometric constraints of perspective transformation are used to determine the starting position of the search when the primitive is located, and the search is only performed within a certain range. If no matching primitive is found within the range, the matching of the to-be-located point is discarded.

Step 5), using the obtained matching point pairs, perform three-dimensional reconstruction based on the principle of stereo vision, and obtain the three-dimensional coordinates of the space points corresponding to the two-dimensional points.

First, use the MATLAB camera calibration toolbox to calibrate the camera internal parameter matrix, make a checkerboard calibration picture as the calibration board, use the camera to shoot the calibration board from different angles, get as many pictures as possible, and measure the checkerboard on the calibration board with a ruler. The actual side length of the small square, the unit is millimeters. Input the obtained data into the calibration toolbox to get the calibration result of the camera's internal parameters.

Secondly, the eight-point algorithm of RANSAC is used to calculate the value of the fundamental matrix, and eight point pairs are randomly selected from the point pairs of the same name, and the result of the fundamental matrix is calculated by solving the linear equation system, and then the support is found in all the original point pairs. For the point pairs of the calculated fundamental matrix, if the number of supported point pairs is large enough, the result of the calculated fundamental matrix is considered to be credible, and all supporting points are used to calculate the final fundamental matrix result by the least squares method. , on the contrary, if only a small number of matching point pairs satisfy the basic matrix result obtained by the initial calculation, repeat the above steps until the optimal solution is found. After the optimal solution of the fundamental matrix is obtained, the fundamental matrix and the camera internal parameter matrix are operated to obtain the result of the essential matrix.

Again, the essential matrix is only related to the relative position and attitude of the camera when taking two pictures. The essential matrix is decomposed by SVD to obtain the transformation matrix [R|t] representing the right image relative to the left image, and four results are obtained, There is only one of these four results, so that the points calculated by the triangulation principle all fall in front of the two cameras, and the unique result can be obtained by taking the point calculation and verifying.

Finally, using the triangulation method, the obtained [R|t] matrix is used to triangulate the point pairs with the same name, and the three-dimensional coordinates of the spatial points corresponding to the two-dimensional points are calculated.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. All within the spirit and principle of the present invention, Any modification, equivalent replacement, improvement, etc. made should be included within the protection scope of the present invention.

Claims (5)

1. A three-dimensional reconstruction method based on texture image coding is characterized by comprising the following steps:

step 1), projecting a color coding image generated by using an M array coding matrix onto an object to be detected, and shooting an area to be detected by using a camera to obtain a left image and a right image;

step 2), extracting a projected target area of the image and the right image based on a Hough circle detection method and a perspective transformation principle so as to avoid the influence of environmental impurities on the projection area;

step 3), respectively enhancing the left image and the right image after the target area is extracted based on a color migration technology, carrying out color identification, converting the color information of the image into a code value, and obtaining a left image color identification image and a right image color identification image;

step 4), directly decoding the color identification image obtained in the step 3) by aiming at the M array coding method to obtain a matching point pair;

the specific implementation manner of the step 4) is as follows;

respectively constructing a right neighborhood map and a lower neighborhood map according to the left image color identification map; the process of constructing the right neighborhood graph is to search right for each pixel point in the color identification graph until the first primitive pixel after the white frame is searched, and record the code value and the coordinate of the point in the right neighborhood graph, and the process of constructing the lower neighborhood graph is to search downwards for each pixel point in the color identification graph until the first primitive pixel after the white frame is searched, and record the code value and the coordinate of the point in the lower neighborhood graph; the method for constructing the right neighborhood graph and the lower neighborhood graph is based on the condition that the pixel point is positioned in the single color block, if the pixel point is positioned on the white frame, the pixel point is searched rightwards or downwards along the point until the pixel point of the first code value of 0, 1 or 2 is obtained through searching, and the point is used as a starting point;

secondly, for an undetermined point in the left image, searching a right element which is most adjacent to the current element in the right neighborhood image and searching a lower element which is most adjacent to the current element in the lower neighborhood image by using a color identification image, a right neighborhood image and a lower neighborhood image, constructing a 3 x 3 window where the current element is located, putting the window into an M array coding matrix corresponding to a color coded image for template matching, and realizing window positioning and element positioning, wherein the element refers to a single color block in the color coded image, the values of the coding matrix correspond to the elements of the color coded image one by one, and the values of the coding matrix are respectively corresponding to the elements of three colors of red, green and blue when the coded image is designed;

thirdly, further accurately positioning the interior of the element according to the position relation between the pixel coordinate of the to-be-positioned pixel and a white frame on the periphery of the element, if the to-be-positioned pixel is positioned in the element, respectively retrieving the to-be-positioned pixel in the upper, lower, left and right directions until the to-be-positioned pixel touches the white frame and stops, namely retrieving a point with a first code value of 3, stopping, calculating the number of pixels passing through the retrieval in the four directions respectively, and realizing positioning through simple calculation; if the undetermined point is not in the primitive, searching rightwards or downwards until the first point in the primitive is searched within the threshold range, and taking the point as a starting point to perform accurate positioning;

finally, searching a matching point of the undetermined point in the right image, positioning the primitive according to the position of the to-be-positioned point in the left image, and then performing further primitive accurate positioning similarly to the steps of primitive positioning and primitive accurate positioning; when the primitive is positioned, the geometric constraint of perspective transformation is firstly utilized to determine the initial position of search, the search is only carried out in a certain range, and if the matched primitive cannot be searched in the range, the matching of the point to be positioned is abandoned;

and 5) carrying out three-dimensional reconstruction based on the stereoscopic vision principle by using the obtained matching point pairs to obtain three-dimensional coordinates of the space points corresponding to the two-dimensional points.

2. The method of claim 1, wherein the texture image coding-based three-dimensional reconstruction method comprises: the specific implementation manner of the step 1) is as follows;

firstly, projecting a designed single color coding image to the surface of an object to be measured through a projector, then shooting images of the object to be measured from different angles by adopting a camera to obtain a left image and a right image, wherein the color coding image is generated by utilizing an M array coding matrix.

3. The method of claim 1, wherein the texture image coding-based three-dimensional reconstruction method comprises: the specific implementation manner of the step 2) is as follows;

firstly, carrying out gradient edge detection on an image to obtain a binary image of edge detection; secondly, detecting a circular function by using Hough transform, selecting three input parameters of proper minimum distance, maximum radius and minimum radius, and detecting to obtain the center coordinates of four circles in the camera-shot image, namely the coordinates of four corner points of the projection area; and finally, taking the four detected corner coordinates as original coordinates, taking the four corner coordinates of the designed color coded image as coordinates after perspective transformation, substituting the coordinates of the four groups of two-dimensional mapping points into a perspective transformation equation set, calculating values of eight unknowns, and obtaining a perspective transformation matrix of the left and right image extraction target areas.

4. The method of claim 1, wherein the texture image coding-based three-dimensional reconstruction method comprises: the specific implementation manner of the step 3) is as follows;

firstly, taking a color coding image as a target image of color migration, taking an image with an extracted target area as an original image of the color migration, converting the target image and the original image from an RGB color space to an l alpha beta color space, and approximating an orthogonal luminance component l and two chrominance components alpha and beta;

secondly, respectively calculating the mean value and the standard deviation of two images in the l alpha beta space, firstly subtracting the mean value of the l alpha beta channel of the target image from the data of the l alpha beta channel of the target image, then scaling the obtained new data according to the proportion, wherein the scaling coefficient is the ratio of the standard deviation of the original image to the standard deviation of the target image, and finally respectively adding the obtained results to the mean value of the l alpha beta channel of the original image to obtain the original image after color migration processing with the same mean value and the same variance as the target image;

finally, carrying out color identification on the image subjected to the color migration treatment; converting the image subjected to color migration processing from the l alpha beta color space to an RGB color space, extracting the RGB value of each pixel, classifying the pixels through a judgment statement, if the gray values of three channels are all larger than a threshold t1, identifying the pixels as white, if the gray value of a red channel is maximum and the gray difference value between the red channel and the other two channels is larger than t2, identifying the pixels as red, wherein the identification principle of the green and the blue is the same as that of the red; and for the identified pixels of the three colors, respectively representing the pixels by corresponding 0, 1 and 2 in the design of a coding pattern, and representing the pixels by corresponding 3, and obtaining a matrix of the image size, namely obtaining the color identification graph.

5. The method of claim 1, wherein the texture image coding-based three-dimensional reconstruction method comprises: the concrete implementation manner of the step 5) is as follows;

firstly, calibrating a camera internal reference matrix by using an MATLAB camera calibration tool box, making a checkerboard calibration picture as a calibration board, shooting the calibration board by using a camera from different angles to obtain pictures as many as possible, taking the actual side length of a checkerboard small square on the calibration board by using a ruler, wherein the unit is millimeter, and inputting the obtained data into the calibration tool box to obtain a calibration result of the camera internal reference;

secondly, calculating the value of a basic matrix by adopting an eight-point algorithm of RANSAC, randomly selecting eight point pairs from the same-name point pairs, calculating by a method for solving a linear equation set to obtain the result of the basic matrix, then searching for the point pairs supporting the calculated basic matrix in all the original point pairs, if the number of the supported point pairs is enough, considering the result of the calculated basic matrix to be credible, and pinching off by using all the supported point pairs through a least square method to obtain a final basic matrix result, otherwise, if only a small number of matched point pairs meet the basic matrix result obtained by initial calculation, repeating the steps until an optimal solution is found; after the optimal solution of the basic matrix is obtained, the basic matrix and the camera internal reference matrix are operated to obtain the result of the essential matrix;

and finally, the essential matrix is only related to the relative position and posture relation of the cameras when two pictures are shot, the essential matrix is subjected to SVD (singular value decomposition) to obtain a transformation matrix [ R | t ] representing the right image relative to the left image, four results are obtained through calculation, only one of the four results enables points obtained through the triangulation principle calculation to fall in front of the two cameras, and the only result is obtained through point taking calculation verification.

Images