CN110766669B - Pipeline measuring method based on multi-view vision - Google Patents

Abstract

The invention discloses a pipeline measurement method based on multi-eye vision. The method consists of four parts: image acquisition and preprocessing, centerline and outer diameter extraction, camera calibration and camera matrix conversion, pipeline centerline matching and outer diameter expansion. The system places the camera on the built measurement frame, performs image acquisition and preprocessing, and obtains a grayscale image of the pipeline. Use the distance transformation and the minimum path method to obtain the pipeline centerline, extract the pipeline centerline feature points and discretize the feature points; calibrate the corresponding binocular cameras (front camera, rear camera), and obtain the The transformation matrix from the side camera coordinate system to the front side camera coordinate system; according to the obtained centerline feature points and camera calibration parameters, a dynamic matching method is used to restore the three-dimensional information of the pipeline centerline, and the previous pictures taken from different perspectives are processed. The obtained outer diameter information is expanded on the three-dimensional pipeline centerline, and finally the spatial position of the pipeline centerline and the corresponding outer diameter information are obtained.

Description

A Multi-Vision-Based Pipeline Measurement Method

technical field

The invention relates to the technical field of visual measurement, and mainly relates to a pipeline measurement method based on multi-eye vision.

Background technique

Piping systems are widely used in nuclear reactors, aviation, aerospace, shipbuilding and automobile industries. They are responsible for the transmission and measurement of gas, liquid and other media. They are an important part of high-end mechanical and electrical products. Security has important implications. Therefore, high-precision pipeline processing is currently a hot issue.

The measurement of the spatial geometry after the pipeline processing mainly adopts the model method, the three-coordinate measuring instrument and the measurement method based on the laser CCD device. The profiling method can only roughly check the shape of the pipeline, but cannot accurately measure the position of the endpoints of the pipeline and the distance between the endpoints. Although the measurement accuracy of the CMM is high, it has strict requirements on the measurement environment and limited measurement range. The measurement method based on the laser CCD device requires workers to move the measuring fork carefully when measuring the endpoint, which is difficult to operate.

In academia, Zhang Tian's paper "Research on Digital Pipeline Measurement Method Based on Multi-ocular Vision" proposed a pipeline reconstruction technology based on centerline matching and machine learning. The machine vision algorithm is used to calculate the coordinates of the pipeline control points, and the pipeline control point coordinates are used to fit the pipeline centerline, and then the three-dimensional model of the pipeline is reconstructed.

Based on the above analysis, the existing technology in the field of pipeline measurement has the following shortcomings: the traditional measurement method of pipeline space geometry is either not accurate enough, or has strict requirements on the measurement environment, or needs to use some high-cost instruments, or It is a method of manual interaction, which has high cost and low efficiency. The difference from the method proposed by Zhang Tian is as follows: 1. In Zhang Tian's paper, the method of image science is used in the extraction of the center line, but this method cannot guarantee that the exact center line is extracted. More efficient distance transformation and minimum path method; 2. In the centerline matching link, Zhang Tian's paper adopts the NURBS curve fitting method, and the present invention proposes a direct calculation method for dynamic matching, which is more concise; The invention proposes an edge detection method based on centroid transformation to calculate the outer diameter of the pipeline.

SUMMARY OF THE INVENTION

Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a pipeline measurement method and system based on multi-eye vision to solve the problems of complex process, high cost and low precision in the current pipeline measurement technology. In this method, the pipeline reconstruction technology based on centerline matching can be used to identify the pipeline path and the outer diameter with high precision, and a complete 3D model of the pipeline can be obtained by cooperating with the modeling software.

Technical scheme: In order to realize the above-mentioned purpose, the technical scheme adopted in the present invention is:

1. A pipeline measurement method based on multi-eye vision, comprising the following steps:

(1) Image acquisition and preprocessing

Step 1. Build an image acquisition system, and capture pictures through binocular cameras distributed on the front and rear sides of the pipeline to obtain a grayscale image of the pipeline; binarize the grayscale image to make the background and the pipeline Distinguish each other; use contour comparison technology to extract the target area ROI from the binarized image to obtain pipeline contour information;

(2) Extraction of center line and outer diameter

Step 2.1. According to the pipeline outline information, use the Euclidean distance transformation method to calculate the Euclidean distance from each point in the pipeline to the pipeline outline, obtain the distance transformation map of the pipeline, and record the Euclidean distance from each point inside the pipeline outline to the pipeline boundary;

Step 2.2. Search the pipeline boundary through image processing, obtain the pixel coordinates of the endpoints on both sides of the pipeline, and use the Dijkstra minimum path algorithm to obtain the minimum path between the two points with the endpoint on one side of the centerline as the starting point and the endpoint on the other side as the endpoint. the centerline of the pipe;

Step 2.3. Use the local feature point discrete method to reduce the eigenvalue, regard the center line as a function, and the required feature point is a part of the stagnation point of the function, optimize all the stagnation points, and obtain the center line feature point set;

Step 2.4, adopt the edge detection method based on the centroid of the shape, obtain and store the size of the outer diameter corresponding to each feature point in the centerline feature point set;

(3) Camera calibration and multi-camera matrix conversion

Step 3.1. Obtain the external parameters composed of the rotation matrix and the translation matrix and the internal parameters of the camera through the camera calibration, and obtain the matrix of the transformation from the pipeline coordinates in the world coordinate system to the pixel coordinates of the camera window according to the internal and external parameters;

Step 3.2. Convert the camera coordinate system according to the external parameter information of the camera, use the same calibration plate to calibrate the rear camera and the front camera at the same time, and obtain the rotation matrix required for the transformation of the rear camera coordinate system to the front camera coordinate system and translation matrix;

(4) Pipe centerline matching and outer diameter expansion

Step 4.1. Perform viewport transformation recovery, perspective projection transformation recovery, view transformation recovery, and model transformation recovery on the centerline feature points obtained by the front and rear cameras, respectively, to obtain the front and rear camera centers pointing forward, and the rear camera imaging Two sets of ray families of the plane;

Step 4.2, according to the rotation matrix and translation matrix obtained in step 3.2, convert the ray family with the center of the rear camera pointing to the imaging plane of the rear camera into the ray family in the front camera coordinate system;

Step 4.3. Use the dynamic matching method. According to the centerline endpoints obtained in step 2.2, start matching from one side endpoint. According to this dynamic method, match until the other side endpoints in turn to obtain the spatial position of the pipeline centerline. line segment representation;

Step 4.4. Expand the outer diameter of the centerline feature point under each view to obtain the outer diameter information of all directions of the pipeline to be tested; adopt the curve fitting method based on the least squares method, and regard the outer diameter at the feature point as The function value, the distance between the feature point and the end point is regarded as an independent variable, a polynomial curve fitting is performed on this series of points, and the fitting curve is selected according to the principle of the smallest sum of deviation squares, and the final function obtained is the outer diameter at different distances from the end point. length;

The pipeline to be tested can be reconstructed according to the spatial coordinates and outer diameter information of the centerline of the pipeline to be tested in the front camera coordinate system.

Further, the edge detection method based on the shape centroid described in the step 2.4 is as follows:

Without considering noise, the one-dimensional mathematical model of image information is simply expressed as:

f(x)=u(x)*g(x)

Among them: u(x) is the original ideal signal, f(x) is the one-dimensional image information; g(x) is the point spread function, which is generally approximated as a Gaussian function:

Set u(x) as ideal step edge, ideal impulse edge, and ideal roof edge respectively. After the point spread function, the sharp edge is smoothed into a fuzzy edge, and the symmetry analysis of the derivative is carried out to obtain the calculation formula of the boundary. for:

Discretize the formula, f _i is the sampling value of f(x) at x _i , and the differential value at x _i is replaced by the mean of the forward difference and the backward difference here as follows:

The formula can be changed to:

Calculate the difference matrix, use the row difference template and the column difference template to convolve the grayscale images of the pipeline respectively, and record the resulting matrices D ₁ , D ₂ ; select the difference threshold T according to the statistical characteristics of the pipeline matrix; If the elements in D ₁ , D ₂ are smaller than T, set to 0, and the non-zero continuous interval of the two matrices is the edge transition interval; use the discretization formula to calculate the value of the edge source point and store it into the edge of the gray image; according to the center line feature point point Set, the outer diameter is the number of pixels obtained in the distance transformation plus the edge point value calculated earlier.

Further, the dynamic matching strategy in the step 4.3 is as follows:

According to the endpoint information obtained from the image processing link, start matching from the endpoint on one side, record the ray corresponding to this endpoint, and use this as the reference ray;

According to the principle of binocular imaging, select the reference ray and the next ray with the center of the front camera pointing to the projection point of the imaging plane of the front camera in the front camera coordinate system to form a plane, and select the center of the rear camera to point to the rear side in this coordinate system The reference ray of the projection point of the camera imaging plane and the next ray form another plane, and the intersection of the two planes is obtained; the intersection and the four rays form four intersections, and the intersection of these two rays and the intersection is selected according to the previously recorded reference rays. As the reference point, the first point far from this intersection is the feature point of the pipeline center line to be tested obtained in this match, and this point is used as the reference point in the next match, and the ray family corresponding to this point is backward in the next match Select a ray; after matching, the spatial feature points of the pipeline center line to be tested are obtained, and the center line of the pipeline to be tested can be obtained by connecting

Beneficial effects: This method proposes a pipeline centerline extraction method based on path transformation and a pipeline centerline reconstruction measurement method based on pipeline centerline matching, which are widely used in the finishing and measurement of pipeline parts. At the same time, a sub-pixel detection method based on centroid transformation is proposed for the measurement of the outer diameter of the pipeline. The method can effectively solve the problems of complex measurement process, high measurement cost, low measurement accuracy, and difficult pipeline model reconstruction in the current pipeline measurement application, and obtains the pipeline centerline polyline family and the corresponding outer diameter information, which can be completely to indicate all the information of the pipeline.

Description of drawings

Fig. 1 is a schematic diagram of each structural relationship of the system of the present invention

2 is a schematic diagram of the construction of the system framework of the present invention;

Fig. 3 is a schematic diagram of the conversion process from the pipeline to be tested to the captured picture;

Fig. 4 is the schematic diagram of perspective projection principle;

Fig. 5 is the schematic diagram of the binocular imaging principle of the present invention;

FIG. 6 is a flow chart of the dynamic matching strategy of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

The multi-vision-based pipeline measurement method provided by the present invention mainly includes (1) image acquisition and preprocessing; (2) centerline and outer diameter extraction; (3) camera calibration and camera matrix conversion; (4) pipeline centerline Matching and OD expansion.

A specific embodiment is given below

(1) Image acquisition and preprocessing

First build an image acquisition system, use a steel frame with a length of about 1.5m as the frame of the image acquisition system, and lay a light-emitting board at the bottom of the frame to eliminate the shadow caused by the lighting of the high-level light source and simplify the pipeline background in image processing. operate. If the desktop is used as the stage, the texture information on the desktop will cause great trouble, and the strong light of laying the light-emitting board can make the background of the pipe picture close to white.

According to the accuracy analysis of the 3D reconstruction simulation of the ideal polyline segment, the image acquisition module needs to use a camera with at least 5 million pixels. Place the cameras in the system frame, as shown in Figure 1-2, and place one camera in the front and the rear to form a pair of camera groups. Take a picture of the pipeline to be measured placed on the luminous plate. The camera uses python and opencv for image capture. Because the subsequent image processing does not require color information, the camera parameters need to be adjusted, and the goal is to obtain the original image as close to the grayscale image as possible.

Convert the captured pipeline image into a grayscale image, perform binarization processing on the grayscale image, and select an appropriate threshold to distinguish the background from the pipeline and some other boundaries. The ROI area is extracted for the second picture, and the contour comparison technology is used. Since the camera is aligned with the pipeline when it is placed, the largest contour in the captured picture is the contour of the pipeline. Use opencv to extract all graphic contours contained in the image, eliminate small contours, and retain the largest contour, that is, to complete the ROI area extraction.

(2) Extraction of center line and outer diameter

Distance transformation refers to calculating the shortest distance from each point inside the object to the object boundary. The centerline of a pipe is the set of points furthest from the pipe boundary. Distance transformation includes Euclidean distance transformation and non-Euclidean distance transformation. Non-Euclidean distance transformation has the characteristics of low complexity and high efficiency, and can obtain similar results, while Euclidean distance transformation has higher complexity and can obtain accurate results. The present invention uses the Euclidean distance transform. Through the Euclidean distance transformation, a scalar field can be obtained to record the Euclidean distance from each point inside the pipeline outline to the pipeline boundary.

Since the pipeline centerline is the set of points farthest from the boundary inside the pipeline, after obtaining the Euclidean distance from each point inside the pipeline to the boundary, the centerline extraction problem becomes the minimum distance from the endpoint on one side of the pipeline to the endpoint on the other side. Path selection problem. The endpoints on both sides of the pipeline are obtained by graphical methods. The specific operation is to continuously refine the pipeline, identify and remove the outermost outline of the pipeline until only a single pixel remains in the pipeline. This is the endpoint of the single-pixel curve on both sides of the pipeline. endpoint. It should be noted that there is a certain error between this single-pixel curve and the pipeline center line, because the contour is refined around the contour operation, so there will be an error of one to two pixels with the real center line.

After getting the coordinates of the endpoints on both sides of the pipe. Select one end point as the starting point of the path. It should be noted that since the two cameras are placed opposite to each other, the captured images are symmetrical. Therefore, if the front camera selects one end point as the starting point of the path, the back side camera must select the other end point. The endpoints can be distinguished by the size of the sum of the horizontal and vertical coordinates of the pixels. After selecting the starting point, the other end point is the end point.

The pipe centerline is the minimum path between the start and end points. The present invention adopts Dijkstra minimum path algorithm. The specific process is: find the starting point and the adjacent shortest path; choose another path, compare the ones that also go to this node, update the path with the least cost, and update the path; repeat this step until the last node; calculate the final path.

After the center line is obtained, the feature points are discretized. The center line is regarded as a function, and the feature points of the center line are some stagnation points of the function. The stagnation point screening strategy is to calculate the distance from a feature point to the two ends of the curve, and the two lines formed by this point and the two endpoints The sum of the distances is compared, and if it reaches a certain level, it will be discarded, otherwise it will be retained. Get the discretized centerline feature point set.

Among them, for the outer diameter of the pipeline, the present invention adopts an edge detection method based on the centroid of the shape. After a detailed analysis of the characteristics of the three basic types of edges: the step edge, the impulse edge and the ridge edge, it is found that the first-order derivatives of these three basic edge types are symmetric functions about the edge points, and their absolute values are An even symmetric function about the edge point. From the point of view of the graph, this symmetric point is just the centroid of the graph. Therefore, the sub-pixel edge detection problem can be transformed into the problem of solving the graph centroid. It is generally believed that the image is formed by the original ideal information under the convolution of the point spread function. Without considering the influence of noise, its one-dimensional mathematical model is simply expressed as:

f(x)=u(x)*g(x)

Among them: u(x) is the original ideal signal, f(x) is the one-dimensional image information; g(x) is the point spread function, which is generally approximated as a Gaussian function:

Set u(x) as the ideal step edge, ideal pulse edge, and ideal roof edge respectively. After the point spread function, the sharp edge is smoothed into a fuzzy edge. After the symmetry analysis of the derivative, the calculation of the boundary is obtained. The formula is:

In order to facilitate the computer to implement this algorithm, it is necessary to discretize the formula. Let f _i be the sampling value of f(x) at x _i , and the differential value at x _i is replaced by the mean value of the forward difference and the backward difference here. , denoted as:

The formula can be changed to:

The specific process of realizing this algorithm is: calculate the difference matrix, use the row difference template and the column difference template to convolve with the pipeline grayscale image, and record the resulting matrices D ₁ , D ₂ ; select the difference threshold T according to the statistical characteristics of the pipeline matrix ;Calculate and select the calculation interval, set 0 for the elements in the matrix D ₁ , D ₂ less than T, the non-zero continuous interval of the two matrices is the edge transition interval; use the discretization formula to calculate the value of the edge source point and store it in the grayscale Image edge; according to the centerline feature point set, the outer diameter is the number of pixels obtained in the distance transformation plus the edge point value calculated earlier.

(3) Camera calibration and multi-camera matrix conversion

Calibrate both cameras. Camera calibration is one of the basic problems of computer vision, and it is the premise of obtaining three-dimensional information from two-dimensional images. Calibration is to use the known information in the two-dimensional image captured by the camera to be calibrated to obtain all the unknown parameters in the camera imaging model, including the internal parameters representing the internal structure of the camera and the external parameters of the camera's spatial pose.

Camera calibration methods are divided into the following four categories: ① calibration methods using three-dimensional calibration objects; ② calibration methods using two-dimensional calibration objects; ③ calibration methods using one-dimensional calibration objects; ④ self-calibration methods. The calibration method using two-dimensional calibration objects has high precision and simple operation, and has gradually become a commonly used calibration method in practical applications. The process of camera calibration is to optimize the process of solving the internal and external parameters of the camera by solving the projection equation. Based on the idea of the two-dimensional plane calibration method, after the multi-cameras are grouped, the binocular stereo cameras are calibrated at the same time.

The two-dimensional plane calibration method requires the use of a calibration plate. The form and production quality of the calibration plate will affect the calibration accuracy of the camera. The calibration plate mainly includes a checkerboard calibration plate and a circular calibration plate. The present invention uses a checkerboard calibration plate. The checkerboard calibration board uses the grid right-angle points as the calibration points, and uses python to capture pictures. The two cameras each capture 15-20 pictures, and try to make the checkerboard cover as large an area as possible. There are different angles. The calibration tool of the image processing package, such as the built-in calibration tool of Matlab, to calibrate the camera, the following parameters can be obtained: the rotation matrix of the camera, the translation matrix, and the internal parameter matrix.

The matrix conversion between cameras is performed according to the camera external parameter information, and the same calibration board is used to calibrate the two cameras at the same time, so that the two cameras have the same world coordinate system when they are calibrated. For a certain point P on the calibration disk (under the world coordinate system), it is transformed into P _t in the rear camera coordinate system through the external parameter matrix of the rear camera, and is transformed into the front camera through the external parameter matrix of the front camera. _Pr in the side camera coordinate system. P _t , P _r can be transformed by a rotation matrix R and a translation matrix T:

P _r =RP _t +T

The rotation matrix expression is:

为后侧相机单独标定时得到的旋转矩阵求逆R _r is the rotation matrix obtained when the front camera is independently calibrated,

Invert the rotation matrix obtained when the rear camera is individually calibrated

The translation matrix expression is:

_T =Tr-RT _t

T _r is the translation matrix when the front camera is individually calibrated, and T _t is the translation matrix when the rear camera is independently calibrated.

(4) Pipe centerline matching and outer diameter expansion

After image acquisition and preprocessing, centerline extraction and discretization, the pixel position information of feature points under the camera view is obtained. After camera calibration, the matrix parameters and perspective projection matrix information of the transformation from the world coordinate system to the camera coordinate system are obtained. According to the camera parameters, the conversion relationship from the imaging plane point to the picture pixel point is obtained. As shown in Figure 3, each point on the pipeline to be tested undergoes the transformation shown in Figure 3 to become a pixel in the image of the camera. The purpose of this module is to use the principle of stereo imaging to restore the position information of points in the world coordinate system from pixel points.

Perform reverse view transformation for each point in the feature point set of the pipeline picture obtained by the rear camera, and restore the pixel information to the two-dimensional point coordinates on the camera imaging surface under the camera coordinate system according to the picture size and starting position. Since the depth information is discarded in perspective projection, as shown in Figure 4, any point (such as point M, point N) on the line passing through the center eye of the camera and point P on the imaging plane will be projected to point P on the imaging plane, After the view transformation, the same pixel will be obtained. Therefore, restoring the perspective projection cannot restore the original point information, and can only point to a ray of point P on the imaging surface from the center eye of the camera, denoted as L _i . It is represented by two points (eye, P) in the rear camera coordinate system.

For each feature point of the front camera, a ray pointing from the center of the camera to a point on the imaging surface (the point obtained by the reverse view transformation) can also be obtained, but this ray is represented in the coordinate system of the front camera. The camera coordinate system conversion relationship obtained by the calibration module converts the ray in the rear camera coordinate system obtained in the previous step into a ray in the front camera coordinate system. This operation is applied to all the feature points on the center line of the images captured by the two cameras, and two sets of rays in the front camera coordinate system are obtained, which are the ray family of the front camera center pointing to the projection point of the front camera imaging plane and The ray family with the center of the rear camera pointing to the projection point of the imaging plane of the rear camera.

Using the principle of binocular imaging, as shown in Figure 5, two rays can determine a plane, and the centerline of the pipeline is located in the intersection of the two planes. Specifically, it is the part between two points among the four intersection points generated by the four rays and the intersection line. The solution to determining which two points are related to the matching strategy. A dynamic matching method is adopted. According to the endpoint information obtained from the image processing, the matching is performed from one endpoint, and the ray corresponding to this endpoint is recorded, which is used as the reference ray. The matching method is based on the principle of binocular imaging, select the reference ray and the next ray that point to the projection point of the front camera imaging plane from the front camera center in the front camera coordinate system to form a plane, and select the rear camera center in this coordinate system. The reference ray and the next ray pointing to the projection point of the imaging plane of the rear camera form another plane, and the two planes are intersected. The intersection line and the four rays form four intersection points. According to the previously recorded reference rays, the intersection point of these two rays and the intersection line is selected as the reference point. The first point far from this intersection point is the centerline feature of the pipeline to be tested obtained by this match. point. This point is used as the reference point for the next matching, and a ray is selected backward in the next matching of the ray family corresponding to this point, as shown in Figure 6. After matching, the spatial feature points of the center line of the pipeline to be tested are obtained, and the center line of the pipeline to be tested can be obtained by connecting.

Expand the outer diameter of the centerline feature point under each view to obtain the outer diameter information of the pipeline to be tested in all directions. A curve fitting method based on the least squares method was used. The outer diameter at the feature point is regarded as a function value, and the distance from the feature point to the end point is regarded as an independent variable, and a polynomial curve fitting is performed on this series of points. The fitting curve is selected according to the principle of the smallest sum of deviation squares, and the final function represents the length of the outer diameter at different distances from the end point.

The pipeline to be tested can be reconstructed according to the spatial coordinates of the centerline of the pipeline to be tested (under the coordinate system of the front camera) and the outer diameter information.

The above is only the preferred embodiment of the present invention, it should be pointed out that: for those skilled in the art, without departing from the principle of the present invention, several improvements and modifications can also be made, and these improvements and modifications are also It should be regarded as the protection scope of the present invention.

Claims (2)

1. a pipeline measurement method based on multi-eye vision, is characterized in that: comprise the following steps: (1) Image acquisition and preprocessing Step 1. Build an image acquisition system, capture pictures through binocular cameras distributed on the front and rear sides of the pipeline, and obtain a grayscale image of the pipeline; binarize the grayscale image to make the background and the pipeline Distinguish each other; use contour comparison technology to extract the target area ROI from the binarized image to obtain pipeline contour information; (2) Extraction of center line and outer diameter Step 2.1. According to the pipeline outline information, use the Euclidean distance transformation method to calculate the Euclidean distance from each point in the pipeline to the pipeline outline, obtain the distance transformation map of the pipeline, and record the Euclidean distance from each point inside the pipeline outline to the pipeline boundary; Step 2.2. Search the pipeline boundary through image processing, obtain the pixel coordinates of the endpoints on both sides of the pipeline, and use the Dijkstra minimum path algorithm to obtain the minimum path between the two points with the endpoint on one side of the centerline as the starting point and the endpoint on the other side as the endpoint. That is, the center line of the pipeline, recorded as the pipeline; Step 2.3. Use the local feature point discrete method to reduce the eigenvalue, regard the center line as a function, and the required feature point is a part of the stagnation point of the function, optimize all the stagnation points, and obtain the center line feature point set; Step 2.4, adopt the edge detection method based on the centroid of the shape, obtain and store the size of the outer diameter corresponding to each feature point in the centerline feature point set; (3) Camera calibration and multi-camera matrix conversion Step 3.1. Obtain the external parameters composed of the rotation matrix and the translation matrix and the internal parameters of the camera through the camera calibration, and obtain the matrix of the transformation from the pipeline coordinates in the world coordinate system to the pixel coordinates of the camera window according to the internal and external parameters; Step 3.2. Convert the camera coordinate system according to the external parameter information of the camera, use the same calibration plate to calibrate the rear camera and the front camera at the same time, and obtain the rotation matrix required for the transformation of the rear camera coordinate system to the front camera coordinate system and translation matrix; (4) Pipe centerline matching and outer diameter expansion Step 4.1. Perform viewport transformation recovery, perspective projection transformation recovery, view transformation recovery, and model transformation recovery on the centerline feature points obtained by the front and rear cameras, respectively, to obtain the front and rear camera centers pointing forward, and the rear camera imaging Two sets of ray families of the plane; Step 4.2, according to the rotation matrix and translation matrix obtained in step 3.2, convert the ray family with the center of the rear camera pointing to the imaging plane of the rear camera into the ray family in the front camera coordinate system; Step 4.3. Use the dynamic matching method. According to the center line endpoints obtained in step 2.2, start matching from one side endpoint. According to this dynamic method, match sequentially to the other side endpoint to obtain the spatial position of the pipeline center line. Line segment representation; specifically, According to the endpoint information obtained from the image processing link, start matching from the endpoint on one side, record the ray corresponding to this endpoint, and use this as the reference ray; According to the principle of binocular imaging, select the reference ray and the next ray with the center of the front camera pointing to the projection point of the imaging plane of the front camera in the front camera coordinate system to form a plane, and select the center of the rear camera to point to the rear side in this coordinate system The reference ray of the projection point of the camera imaging plane and the next ray form another plane, and the intersection line of the two planes is obtained; the intersection line and the four rays form four intersection points, and the intersection point of these two rays and the intersection line is selected according to the previously recorded reference rays. As the reference point, the first point far from this intersection is the feature point of the pipeline centerline obtained by this match, and this point is used as the reference point for the next match, and the ray family corresponding to this point is backward in the next match Select a ray; after matching, the spatial feature points of the center line of the pipeline to be tested are obtained, and the center line of the pipeline to be tested can be obtained by connecting; Step 4.4. Expand the outer diameter of the centerline feature point under each view to obtain the outer diameter information of all directions of the pipeline to be tested; adopt the curve fitting method based on the least squares method, and regard the outer diameter at the feature point as The function value, the distance between the feature point and the end point is regarded as an independent variable, a polynomial curve fitting is performed on this series of points, and the fitting curve is selected according to the principle of the smallest sum of deviation squares, and the final function obtained is the outer diameter at different distances from the end point. length; The pipeline to be tested can be reconstructed according to the spatial coordinates and outer diameter information of the centerline of the pipeline to be tested in the front camera coordinate system. 2. a kind of pipeline measurement method based on multi-eye vision according to claim 1 is characterized in that: the edge detection method based on shape centroid described in described step 2.4 is as follows: Without considering noise, the one-dimensional mathematical model of image information is simply expressed as: f(x)=u(x)*g(x) Among them: u(x) is the original ideal signal, f(x) is the one-dimensional image information; g(x) is the point spread function, which is generally approximated as a Gaussian function:

Set u(x) as ideal step edge, ideal impulse edge, and ideal roof edge respectively. After the point spread function, the sharp edge is smoothed into a fuzzy edge, and the symmetry analysis of the derivative is carried out to obtain the calculation formula of the boundary. for:

Discretize the formula, f _i is the sampling value of f(x) at x _i , and the differential value at x _i is replaced by the mean of the forward difference and the backward difference here as follows:

The formula can be changed to:

Calculate the difference matrix, use the row difference template and the column difference template to convolve the grayscale images of the pipeline respectively, and record the resulting matrices D ₁ , D ₂ ; select the difference threshold T according to the statistical characteristics of the pipeline matrix; If the elements in D ₁ , D ₂ are smaller than T, set to 0, and the non-zero continuous interval of the two matrices is the edge transition interval; use the discretization formula to calculate the value of the edge source point and store it into the edge of the gray image; according to the center line feature point point Set, the outer diameter is the number of pixels obtained in the distance transformation plus the edge point value calculated earlier.

Images