CN108335332A - A kind of axial workpiece central axes measurement method based on binocular vision - Google Patents

Abstract

The axial workpiece central axes measurement method based on binocular vision that the invention discloses a kind of.Monocular calibration is carried out to the left and right cameras of binocular vision system respectively, obtains internal reference, outer ginseng and the distortion parameter of left and right cameras；Stereo calibration is carried out to binocular vision system, obtains the relative position relation of left and right cameras, the parameter iteration that monocular is demarcated is optimized using relative position relation；Axial workpiece is placed in the public visual field of left and right cameras, is taken pictures to axial workpiece by left and right cameras and obtains the two images of left images respectively, to two images correction process；Processing is carried out to the left images after correction and obtains respective axis profile；The central axes of axial workpiece are obtained using the axis profile processing of left images.The method of the present invention, can effectively seek the central axes of axial workpiece, high certainty of measurement, as a result reliably；And can test to the central axes measurement result of space axial workpiece, the monitor station of high price is avoided, cost is reduced.

Description

A method for measuring the central axis of shaft parts based on binocular vision

technical field

The invention belongs to the field of advanced measurement technology, and relates to a binocular vision-based measurement method for the central axis of shaft parts, and is especially suitable for non-contact industrial detection and vision-based robot navigation systems.

Background technique

Shaft parts can be seen everywhere in industrial applications, and the accurate measurement of the three-dimensional pose of shaft parts is of great significance in industrial inspection. Especially in modern manufacturing technology, automated assembly technology has gradually replaced manual complex assembly operations to improve efficiency and ensure product quality and stability. In the process of parts assembly, the shaft hole assembly is a frequently encountered working situation. When the manipulator grabs the shaft parts and assembles them, it is necessary to accurately measure the three-dimensional pose of the central axis of the shaft parts in order to accurately carry out the shaft hole. Assembly, and avoid defects caused by collisions between parts, and then successfully complete the assembly task.

With the development of machine vision technology, relevant research and application tests have been carried out at home and abroad, and the machine vision system is introduced into the shaft hole size measurement. However, most of the current research focuses on the recognition of spatial round holes and their elliptical holes. Compared with parameter measurement, there are relatively few studies on the size of shaft parts and their three-dimensional pose measurement. Cui Yanping and others have studied a method for measuring the three-dimensional attitude of the revolving object based on binocular vision. This method can realize the measurement of the three-dimensional attitude of the revolving object without matching feature points. Due to high reflectivity and uneven illumination in practical applications, using this method will lead to a decrease in accuracy when extracting the sub-pixel straight line equation of the image plane bus, and the verification process of this method requires a precision work turntable, which is expensive. Sun et al. proposed a shaft measurement method based on image processing in the article Shaft diameter measurement using a digital image. This measurement method based on the principle of monocular vision and departing from the principle of stereo geometry must rely too much on image quality and its processing effect. Good stability, and this method needs to use the known diameter of the shaft size as prior knowledge for recalibration in order to achieve higher accuracy, and the application has certain limitations. In addition, the target workpiece axis calculation method proposed by Zhang Kai of Xi'an University of Technology needs to use gray value threshold and RGB color threshold in the image processing process, so it may not be recognized when ordinary black and white cameras are used, and even the calculation fails.

To sum up, it is of great application value to propose a low-cost method that can accurately measure the three-dimensional attitude of the axis of shaft parts without manual intervention.

Contents of the invention

In order to solve the problems existing in the background technology, the present invention proposes a method for measuring the central axis of shaft parts based on binocular vision, which can effectively measure the central axis of space shaft parts with high precision and fast speed, and has been verified by inspection. The method is simple and practical, which greatly saves the experimental cost and improves the efficiency, and provides a new idea for the verification of the binocular vision algorithm.

In order to achieve the above object, the technical solution adopted in the present invention comprises the following steps:

1) According to the Zhang Zhengyou calibration method, the left and right cameras of the binocular vision system are single-targeted, and the binocular vision system is composed of the left and right cameras, and the internal parameters, external parameters and radial distortion parameters of the left and right cameras are obtained; according to the binocular vision stereo calibration method Calculate the structural parameters of the binocular vision system, that is, the relative positional relationship between the left and right cameras: then use the relative positional relationship to optimize the internal parameters, external parameters, and radial distortion parameters of the left and right cameras, and then calculate the left and right The projection matrix of the camera;

2) Put the shaft parts in the public field of view of the left and right cameras, take pictures of the shaft parts through the left and right cameras to obtain two images of the left and right images (left image and right image), and perform Gaussian filtering and graying on the two images in turn. Degree conversion and threshold value processing, then use the radial distortion parameters of the left and right cameras obtained in step 1) to correct the left and right images;

3) Process the corrected left and right images to obtain respective axis profiles;

4) Use the shaft contour processing of the left and right images to obtain the central axis of the shaft part.

The internal parameters, external parameters, radial distortion parameters and projection matrices of the left and right cameras obtained in step 1) include the internal parameter matrix A _l of the left camera, the external parameter matrices R _l and T _l of the left camera, the radial direction of the left camera Distortion parameter matrix K _l , projection matrix M _l of the left camera, intrinsic parameter matrix A _r of the right camera, extrinsic parameter matrices R _r and T _r of the right camera, radial distortion parameter matrix K _r of the right camera and projection of the right camera Matrix M _r .

The relative positional relationship between the left and right cameras obtained in step 1) includes a rotation matrix R and a translation vector T.

In said step 1), the internal reference, external reference and radial distortion parameters of the left and right cameras are optimized using the relative positional relationship, and then the internal reference calculations of the left and right cameras are used to obtain the external parameters of the left and right cameras, specifically:

Using the rotation matrix R and the translation vector T, optimize the internal parameters, external parameters and radial distortion parameters obtained by each camera calibration according to the 3D constraint optimization method to improve the calibration accuracy, and then use the optimized internal parameters and external parameters to calculate the left camera The projection matrix M _l of the right camera and the projection matrix M _r of the right camera.

The 3D constrained optimization method adopts the calculation method proposed by Yi Cui in Precise calibration of binocular visionsystem used for vision measurement, pages 8-10.

The step 3) is specifically: carry out contour detection to the corrected left and right images and the template image, match all the contours in the detected left and right images with the template contours obtained by template image detection, and obtain the template contours in the left and right images respectively. The contour matched by the contour, and used as the shaft contour, that is, the contour corresponding to the shaft part;

There is one and only one rectangular frame matching the shaft part in the template image. The size and shape of the rectangular frame match the shaft part, and the specific implementation is that the rectangular frame is placed in the middle of the template image.

Described step 4) specifically is:

4.1) Approximate the shaft contours in the left and right images with the circumscribed rectangle of the minimum rotation, and extract the long side of each shaft contour. The long sides of the two shaft contours in the left and right images have four long sides _lab , l _cd , l _gh , l _ij , l _ab , l _cd represent the two long sides of the axis profile in the left image respectively, l _gh , l _ij respectively represent the two long sides of the axis profile in the right image;

4.2) Use the following formula to calculate the four space planes S _OAB , S _OCD , S _0′GH , S _0′IJ that pass through the center of the left and right cameras and are tangent to the shaft profile of the shaft parts, and then obtain the normal vectors of the four space planes N _OAB , N _OCD , N _0′GH , N _0′IJ :

l _ab M _l S _OAB =0, l _cd M _l S _OCD =0, l _gh M _r S _0′GH =0, l _ij M _r S _0′IJ =0

In the formula, l _ab , l _cd represent the two long sides of the axial contour of the left image respectively, l _gh , l _ij represent the two long sides of the central axis contour of the right image respectively; M _l , M _r represent the projections of the left and right cameras respectively matrix; S _OAB , S _OCD represent the two space planes that pass through the center of the left camera and are tangent to the two long sides of the shaft profile of the shaft part respectively; S _0′GH and S _0′IJ represent the planes passing through the center of the right camera and respectively Two spatial planes tangent to the two long sides of the shaft profile of shaft parts;

4.3) According to the space plane S _OAB , S _OCD and their respective normal vectors N _OAB , N _OCD calculate the space angle bisector plane S _l between the space plane S _OAB and the space plane S _OCD , according to the space plane S _0'GH , S _0'IJ and their respective normal vectors N _0'GH , N _0'IJ calculate the space angle bisector plane S _r between the space plane S _0'GH and the space plane S _0'IJ , take two space angle bisector planes S The intersection line of _l and S _r is used as the central axis of shaft parts.

Described step 4.1) is specifically:

4.1.1) Use the canny operator to perform edge detection on the corrected left and right images and the template image, store all the edges detected in the left image as the left edge set vector_left, and store all the edges detected in the right image as the right edge set vector_right , record the template edge result obtained by template image edge detection as mode_edge;

4.1.2) Traverse each element in the left edge set vector_left and the right edge set vector_right respectively, match each element with the template edge mode_edge according to the template matching principle, calculate the matching degree respectively, and match all the edges of the left image The degree is stored as the left score set vector_LeftScores, and the matching degree of all edges of the right image is stored as the right score set vector_RightScores;

In the specific implementation of this step, the matchTemplate function of the OpenCV tool is used for matching calculation to obtain the matching degree.

4.1.3) Sort the elements in the left score set vector_LeftScores and the right score set vector_RightScores respectively from small to large, and select the contour corresponding to the smallest score in the two sets as the imaging contour of the central axis in the left and right cameras;

4.1.4) Approximating the two imaging contours obtained in step 4.1.3) by using the minimum rotating rectangle method to obtain two circumscribed rectangles rect_left and rect_right that are extremely similar to the contours;

4.1.5) Select the two long sides in the circumscribed rectangle rect_left of the left image as the two long sides l _ab and l _cd of the axis profile of the left image, and select the two long sides in the circumscribed rectangle rect_right of the right image as the right image The two long sides of the axis profile of , calculate the straight line equation of each long side.

In the implementation of the present invention, a new inspection model is also proposed. The radius of the model is known, and the central axis is visualized. The fitting and reconstruction results of the space straight line are compared with the results obtained by the method of the present invention to verify the radius of the method. Validity and accuracy of dimensional measurements.

The beneficial effects that the present invention has are:

1. The method of the present invention can automatically and high-precision measure the spatial three-dimensional posture of the axis of the shaft part through the binocular vision measurement technology, has the advantage of non-contact, and has high application in the occasions that cannot be measured by traditional methods value, especially for non-contact industrial inspection and vision-based robot navigation systems.

2. The method proposed by the present invention to approximate the contour line of shaft parts on the imaging plane of the left and right cameras by using the minimum rotating circumscribed rectangle can compensate for the detection error caused by uneven illumination and high reflectivity of the part surface, and improve the accuracy; self-designed The test model is simple and practical, which avoids expensive precision test benches, saves experimental costs, and provides a new idea for the test of binocular vision-related algorithms.

Description of drawings

Figure 1 is a schematic diagram of the principle of the method of the present invention.

Fig. 2 is a schematic flow chart of the method of the present invention.

Fig. 3 is a schematic diagram of the new model verification method of the embodiment.

Detailed ways

In order to better understand the present invention, the technical solutions of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

Figure 1 shows the binocular stereo vision system. O _l X _l Y _l Z _l and O _r X _r Y _r Z _r are the left and right camera coordinate systems respectively, o _l u _l v _l and o _r u _r v _r are the left and right image coordinate systems in units of pixels, O _w X _w Y _w Z _w is the world coordinate system, where the Z axis points to the origin O _l of the left camera coordinate system. Let the rotation matrix and translation vector between the world coordinate system and the left camera coordinate system be R _o and t _o respectively, write R _o = [r ₁ r ₂ r ₃ ], where the three-dimensional column vector _{r i} represents the first Column i, i=1,2,3. Let the rotation matrix and translation vector between the left and right camera coordinate systems be R and T respectively, then Among them, X _l and X _r are the coordinates of a certain space point under the coordinates of the left and right cameras respectively, and R and T are determined by binocular camera calibration.

In Fig. 1, EF represents the central axis of the shaft part, abcd represents the profile of the shaft part on the imaging plane of the left camera, and ghij represents the contour of the shaft part on the imaging plane of the right camera.

Describe the implementation steps of the inventive method in detail below:

1. For the left and right cameras, use Zhang Zhengyou’s calibration method (A Flexible New Technique for CameraCalibration. Zhengyou Zhang, December, 2, 1998.) to determine the internal parameter matrix A _l of the left camera, the radial distortion parameter K _l of the left camera, Intrinsic parameter matrix A _r of the right camera, radial distortion parameter K _r of the right camera; and calculate the projection matrix M _l corresponding to the left camera and the projection matrix M _r corresponding to the right camera; according to the binocular vision stereo calibration method (Jeans-Yves Bouguet, Camera Calibration Toolbox for Matlab, MRL-Intel Incorp.), calculate the structural parameters of the binocular vision system, that is, the relative positional relationship between the left and right cameras: rotation matrix R and translation vector T. Among them, the form of the internal parameter matrix is:

Among them, α _l and β _l represent the effective focal length of the left camera in the x-axis direction and the effective focal length in the y-axis direction respectively, u _ol and v _ol represent the principal point coordinates of the imaging plane of the left camera, γ represents the tilt parameter of the left camera coordinate axis, ideally α _r and β _r represent the effective focal length of the right camera in the x-axis direction and y-axis direction respectively, u _or and v _or represent the principal point coordinates of the imaging plane of the right camera, and γ _r represents the coordinate axis of the right camera The tilt parameter, ideally 0.

In this example, the camera calibration and calculation results are:

K _l ＝[-0.0114-0.0578-1.3177]

K _r ＝[-0.0233-0.1116-0.4499]

T = [-59.7610 0.4727 0.7066]

2. Place the binocular vision system near the shaft part to be detected, ensure that the shaft part is within the common field of view of the left and right cameras, and at the same time keep the background as simple as possible, and make the axial direction of the shaft part as parallel as possible to the left and right The line connecting the origin of the camera coordinate system is the baseline of the binocular vision system. The left and right cameras are used to shoot the shaft part at the same time, so that an image containing the image of the shaft part is obtained from the left camera, and correspondingly, an image containing the image of the shaft part is also obtained from the right camera. Use the radial distortion parameter K _l of the left camera to correct the distortion of the left image, and obtain the left image without distortion information, denoted as plane _l . At the same time, use the radial distortion parameter K _r of the right camera to correct the distortion of the right image, and obtain the right image without distortion information, denoted as plane _r .

The specific correction process is: for the image on the left, set the coordinates of an image point containing distortion information in the image coordinate system in pixels as Its normalized image coordinates are Their corresponding image point coordinates without distortion information are (u, v) and (x, y) respectively. According to the literature (DC Brown, Close-range camera calibration, Photogram-metric Engineering, 37(8):855-866, 1971), there are:

Use the coordinate change formula:

Among them, K _l is the radial distortion parameter of the left camera, and A _l is the internal parameter matrix of the camera, both of which can be determined by the single-target calibration of the camera.

can get:

Since the above equations are nonlinear equations, in order to simplify the solution process, the above equations can be approximated as

The above two formulas can be used to correct the distortion of each image point on the left and right images, so as to obtain the image plane _l without distortion information. For the image on the right, the rectification method is exactly the same as that of the left image, and will not be repeated here.

3. Perform adaptive thresholding on the corrected left image plane _l and right image plane _r to binarize the image. The specific adaptive threshold is calculated by the following method:

Assume that the variable t takes an integer value within the gray value range (0-255) accordingly (a total of 256 gray values), and divides the left image into two parts, the background and the foreground, and calculates the following two formulas at the same time:

u＝w ₀ *u ₀ +w ₁ *u ₁

g＝w ₀ *(uu ₀ ) ² +w ₁ *(uu ₁ ) ²

Among them, w ₀ is the proportion of image background pixels to the whole image, u ₀ is the average gray level of image background pixels, w ₁ is the proportion of foreground pixels to the whole image, u ₁ is the average gray level of foreground pixels degree, u represents the average gray level of the entire image, and the calculation result g represents the variance of the gray value of the foreground and background images. Comparing the obtained 256 g values, the value of the variable t is the optimal threshold when the g value is the largest, and image binarization is performed according to the optimal threshold.

Then perform contour detection, and match all detected contours with template contours, calculate the matching similarity score, and obtain the contours of the target shaft in the left and right images according to the score; since the axis direction of the shaft parts is basically parallel to the left and right camera coordinate systems From the connection of the origin, it can be known that the image of the axis on the left and right imaging planes is similar to a rectangle, so this paper uses the minimum rotation circumscribed rectangle to approximate the matched axis contours in the left and right images, and calculates the upper axis contours of the left image plane _l and the right image plane _r For the four axial sides _lab , l _cd , l _gh , l _ij , the method of minimum rotating circumscribed rectangle can also make up for the image contour segmentation error caused by the surface highlight and uneven illumination of shaft parts.

The values of _lab , l _cd , l _gh , and l _ij calculated in this example are:

l _ab : [463.23004,-1358.9934,847808]

l _cd : [463.23004,-1358.9935,292788.81]

_lgh : [459.90479,-1310.1235,987763.63]

l _ij : [459.90479,-1310.1235,432691.72]

4. According to the silhouette edges _lab , l _cd , l _gh , l _ij of the four axis directions on the left image plane _l and the right image plane _r , as well as the left camera projection matrix M _l and the right camera projection matrix M _r , calculate through the center of the left and right cameras And 4 planes S _OAB , S _OCD , S _l′GH , S _0′IJ that are tangent to the side of space axis parts.

Its realization principle is (Cui Yanping, Research on 3D Attitude Measurement Method of Revolving Object Space. Journal of Sensing Technology, Jan. 2007.): It is known that the bus bars AB and CD of the space axis parts in Figure 1 are on the imaging plane of the left camera The images of the busbars GH and IJ on the imaging plane of the right camera are gh and ij, and the equations of the planes S _OAB , S _OCD , S _0'GH , and S _0'IJ in the world coordinate system are:

l _ab M _l S _OAB =0, l _cd M _l S _OCD =0, l _gh M _r S _0′GH =0, l _ij M _r S _0′IJ =0

Simultaneously obtain the normal vectors N _OAB , N _OCD , N _0′GH , N _0′IJ of the four space planes.

Calculated in this example

_SOAB : [1744510.4,-5117928.5,224126.41,0]

S _OCD : [1744510.4,-5117929,-330892.88,0]

S _0′GH : [1731987.6,-4933886,403896.97,-1.0351583e+008]

S _0′IJ : [1731987.6,-4933886,-151174.95,-1.0351583e+008]

_NOAB : [0.32235768, -0.94571155, 0.041414984]

N _OCD : [0.32235768, -0.94571155, 0.041414984]

S _0′IJ : [0.33024019,-0.94075006,0.077011526]

S _0′IJ : [0.3310855,-0.94315809,-0.028898496]

5. Through basic geometric principles, the space angle bisector plane S _l of S _OAB and S _OCD can be calculated from the tangent planes S _OAB , S _OCD and their corresponding normal vectors N _OAB , N _OCD ; according to the tangent planes S _0′GH , S _0′IJ and its corresponding normal vectors N _0′GH and N _0′IJ calculate the space angle bisector plane S _r of S _0′IJ and S _0′IJ ; then the intersection line of the two bisector planes S _l and S _r is the central axis of the space axis.

The results of S _l and S _r calculated in this example are:

S _l : [0.64438975,-1.8904679,-0.019666974,0]

S _r : [0.66132569,-1.8839082,0.048113029,-39.525501]

Finally, a test model is designed in this embodiment. As shown in FIG. 3 , the model is designed into two parts, I and II. Part I is an optical axis with a radius of 30mm and a length of 180mm, and its function is to measure the central axis of the optical axis by the method proposed by the present invention. Part II is a quarter of the optical axis, which is coaxial with the optical axis of Part I and has the same radius, with a length of 100mm. This part of the model visualizes the central axis, and the fitting results of the visualized central axis are compared with The method proposed by the present invention compares the measurement results of the optical axis of the first part to verify the effectiveness and accuracy of the method of the present invention.

In the present embodiment, the fitting result of the visible straight line of the inspection model and the measurement results of the present invention are as follows in Table 1 (the space straight line is represented as the intersection line of two space planes):

Table 1

plane 1 plane 2 axis direction vector Fitting result [0.6471,-1.8912,-0.0201,0] [0.6645, -1.8866, 0.0485, -39.6322] [-0.1296,-0.0447,0.0359] algorithm reconstruction [0.6444,-1.8905,-0.0197,0] [0.6613, -1.8839, 0.0481, -39.5255] [-0.1280,-0.0440,0.0362]

From this, it is calculated that the included angle between the fitting central axis and the algorithm reconstruction central axis is 0.54 ^° , and the distance is 1.11mm. It can be seen from this that the shaft parts based on binocular vision and the minimum circumscribed rectangle proposed by the present invention The axis measurement method can achieve high precision, which provides support for the automatic measurement of shaft parts.

Claims (8)

1. a kind of axial workpiece central axes measurement method based on binocular vision, it is characterised in that comprise the steps of：

1) monocular calibration is carried out to the left and right cameras of binocular vision system respectively according to Zhang Zhengyou standardizations, obtains left and right camera shooting The internal reference of machine, outer ginseng and radial distortion parameter；The structure of binocular vision system is calculated according to binocular vision stereo calibration method Parameter, the i.e. relative position relation of left and right cameras：Then utilize relative position relation to the internal reference of left and right cameras, outer ginseng and Radial distortion parameter optimizes, and the projection matrix of left and right cameras is calculated further according to the internal reference of left and right cameras, outer ginseng；

2) axial workpiece is placed in the public visual field of left and right cameras, axial workpiece take pictures point by left and right cameras Not Huo get left images two images (left image and right image), gaussian filtering, gradation conversion are carried out successively to two images And threshold process, the radial distortion parameter of the left and right cameras then obtained using step 1) carry out at correction left images Reason；

3) processing is carried out to the left images after correction and obtains respective axis profile；

4) the axis profile processing of left images is utilized to obtain the central axes of axial workpiece.

2. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 1, it is characterised in that： The internal reference for the left and right cameras that the step 1) obtains, outer ginseng, radial distortion parameter and projection matrix include left video camera Intrinsic Matrix A_l, left video camera outer parameter matrix R_lAnd T_l, left video camera radial distortion parameter matrix K_l, left video camera Projection matrix M_l, right video camera Intrinsic Matrix A_r, right video camera outer parameter matrix R_rAnd T_r, right video camera radial direction Distortion parameter matrix K_rWith the projection matrix M of right video camera_r。

3. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 1, it is characterised in that： The relative position relation for the left and right cameras that the step 1) obtains includes spin matrix R and translation vector T.

4. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 1, it is characterised in that： In the step 1), the internal reference of left and right cameras, outer ginseng and radial distortion parameter are optimized using relative position relation, then The projection matrix for obtaining left and right cameras is calculated with the internal reference of left and right cameras, outer ginseng, specially：

It is the internal reference that each camera calibration is obtained according to 3D constrained optimizations method, outer using spin matrix R and translation vector T Ginseng and radial distortion parameter optimize, then with after optimization internal reference and outer ginseng calculate the projection matrix M of left video camera_lWith The projection matrix M of right video camera_r。

5. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 1, it is characterised in that： The step 3) is specially：To the left images and template image progress contour detecting after correction, in the left and right figure that will be detected As in all profiles matched with the detected template contours of template image, acquisition left images in respectively with template wheel Wide matched profile, and as axis profile.

6. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 5, it is characterised in that： Have one and only one and the matched rectangle frame of axial workpiece in the template image.

7. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 1, it is characterised in that： The step 4) is specially：

4.1) the axis profile in left images is approached with minimum rotation boundary rectangle, extraction obtains the long side of each axis profile, left The long side of two axis profiles is total in right image four long side l_ab、l_cd、l_gh、l_ij, l_ab、l_cdLeft image centre shaft wheel is indicated respectively Two wide long sides, l_gh、l_ijTwo long sides of right image axis profile are indicated respectively；

4.2) following formula is used to calculate four spaces by left and right cameras center and tangent with the axis profile of axial workpiece Plane S_OAB、S_OCD、S_0′GH、S_0′IJ, and then obtain the normal vector N of four space planes_OAB、N_OCD、N_0′GH、N_0′IJ：

l_abM_lS_OAB=0, l_cdM_lS_OCD=0, l_ghM_rS_0′GH=0, l_ijM_rS_0′IJ=0

In formula, l_ab、l_cdTwo long sides of left image axis profile, l are indicated respectively_gh、l_ijRight image axis profile is indicated respectively Two long sides；M_l、M_rThe projection matrix of left and right cameras is indicated respectively；S_OAB、S_OCDIt indicates to distinguish by left camera center respectively Tangent two spaces plane, S with two long sides of axis profile of axial workpiece_0′GH、S_{0 ＇ IJ}It indicates to pass through right camera center respectively The two spaces plane tangent with two long sides of the axis profile of axial workpiece respectively；

4.3) according to space plane S_OAB、S_OCDAnd its respective normal vector N_OAB、N_OCDCalculate space plane S_OABAnd space plane S_OCDBetween space angle-bisecting plane S_l, according to space plane S_0′GH、S_0′IJAnd its respective normal vector N_0′GH、N_0′IJIt calculates empty Between plane S_0′GHWith space plane S_{0 ＇ IJ}Between space angle-bisecting plane S_r, take two spaces angle-bisecting plane S_lWith S_rIntersection Central axes as axial workpiece.

8. a kind of axial workpiece central axes measurement method based on binocular vision according to claim 7, it is characterised in that： The step 4.1) is specially：

4.1.1 it) uses canny operators to left images and template image progress edge detection after correction, left image is examined All edges measured save as left hand edge set vector_left, and all edges that right image detects are saved as right hand edge collection Vector_right is closed, the obtained template edge result of template image edge detection is denoted as mode_edge；

4.1.2 each member in left hand edge set vector_left and right hand edge set vector_right) is traversed respectively Each element is matched with template edge mode_edge according to template matches principle, calculates separately and matched by element respectively Degree, is stored as left score set vector_LeftScores, by all edges of right image by the matching degree at all edges of left image Matching degree be stored as right score set vector_RightScores；

It 4.1.3) respectively will be in left score set vector_LeftScores and right score set vector_RightScores Element according to being ranked up from small to large, and axis is imaged in left and right centered on the corresponding profile of minimum score in two set of selection Image profiles in machine；

4.1.4) using minimum rotation rectangle method to step 4.1.3) in two image profiles obtaining approach, obtain two A boundary rectangle rect_left and rect_right；

4.1.5) choose left image boundary rectangle rect_left in two long sides as left image axis profile it is two long Side l_ab、l_cd, two long sides in the boundary rectangle rect_right of right image are chosen as the two long of the axis profile of right image Side calculates the linear equation of each long side.