CN105261047B

CN105261047B - A method for extracting the center of butt ring based on short-range short-arc segment images

Info

Publication number: CN105261047B
Application number: CN201510568069.9A
Authority: CN
Inventors: 毛晓艳; 何英姿; 魏春岭; 胡海东; 朱志斌; 唐强; 张海博; 徐栓峰; 龚小谨; 江文婷
Original assignee: Beijing Institute of Control Engineering
Current assignee: Beijing Institute of Control Engineering
Priority date: 2015-09-08
Filing date: 2015-09-08
Publication date: 2019-04-09
Anticipated expiration: 2035-09-08
Also published as: CN105261047A

Abstract

A kind of docking ring center extracting method based on short distance short arc segments image of the present invention, step includes: to carry out polar curve correction for the binocular image comprising short arc segments butt joint ring, carries out Threshold segmentation using standard deviation method, anchor ring extraction is carried out using area attribute, then extracts its outer edge；Fine and close reconstruction is carried out to the annulus region extracted using half global matching process, obtains the three-dimensional point cloud under camera coordinates system；The anchor ring normal vector of calculating point cloud simultaneously will be in coordinate projection to the plane vertical with normal vector；Outer edge is distinguished in new plane, carries out the fitting of standard positive round respectively；It according to obtained outer edge center and radius, is constrained using concentric ring, finally obtains the accurate central coordinate of circle of butt joint ring.The present invention directly projects three dimensional point cloud in the space plane just penetrated from the characteristic of butt joint ring, ellipse fitting is reduced to round fitting, while eliminating erroneous point by the constraint condition of inner and outer ring, has the advantages that result is accurate, calculation amount is small.

Description

A kind of docking ring center extracting method based on short distance short arc segments image

Technical field

The present invention relates to the methods of anchor ring extraction and the estimation of the anchor ring center of circle, and in particular to one kind is based on short distance short arc segments The docking ring center extracting method of image.

Background technique

In field of aerospace, butt joint ring is target common in spacecraft, is only carried out to butt joint ring target quasi- True pose estimation just can be carried out accurate docking, to carry out the space tasks such as Material Transportation, aircraft repairing.Wherein anchor ring Detection and the estimation in the anchor ring center of circle are steps crucial in pose estimation.For decades, the pose estimation of butt joint ring is one always A hot issue, scholars propose the method for many pose estimations.

Currently, the pose estimation of spacecraft is carried out mainly for cooperative target, it can is marked in cooperative target Index point is carrying out that pose estimation can be carried out according to index point position in docking operation with aircraft.Flown based on index point The estimation of row device object pose needs to carry out careful mark before vehicle launch, so that Flight Vehicle Design work becomes cumbersome. Method based on mark point is not suitable for the noncooperative target of the non-labels tokens point in space.In addition, this method only needs to obtain mesh The binocular image of aircraft is marked, corresponding camera apparatus is portable, cheap.

Summary of the invention

Technology of the invention solves the problems, such as: overcoming the deficiencies of the prior art and provide a kind of based on short distance short arc segments image Docking ring center extracting method, this method, which can overcome in conventional method, to be needed the hand labeled index point on butt joint ring and consumes The shortcomings that a large amount of manpowers, while having avoided space annulus and having projected the distortion of projection after two-dimensional surface, do not take pains solution it is oval or The fitting problems of conic section, but the measurement image obtained by binocular camera, restore the three-dimensional information at its anchor ring and edge And project it on the projection plane just penetrated, simplify the positive round for being projected as two-dimensional surface of space annulus, simplify and calculate, only needs Anchor ring center can be accurately estimated according to the topography of annulus, under the premise of guaranteeing certain robustness, there is estimation knot The advantage that fruit is accurate, time-consuming is low.

The technical solution adopted by the present invention is that: as shown in Figure 1

(1) binocular image comprising short arc segments butt joint ring target obtained for short distance carries out polar curve correction, then adopts Binarization threshold segmentation is carried out with standard deviation method, the bianry image after being divided.Utilize the area attribute in bianry image Annulus region extraction is carried out, annulus region uses its outer edge of canny operator extraction after extracting；

(2) fine and close reconstruction is carried out to the annulus region extracted using half global matching process, obtained a large amount of on anchor ring The parallax value of point (being exactly all pixels that can be registrated on anchor ring, determined by the imaging area size and registration rate of anchor ring), To carry out the three-dimensional reconstruction under camera coordinates system, three-dimensional point cloud is obtained；

(3) multi-point fitting is used to the three-dimensional point cloud under camera coordinates system, the method for largest optimization calculates anchor ring normal vector And the three-dimensional point cloud indicated under camera coordinates system is projected in the plane vertical with normal vector；

(4) inward flange and outer edge are distinguished in the plane vertical with normal plane, using multi-point fitting, as a result optimal side Method carries out standard positive round fitting respectively, obtains center and the radius of butt joint ring outer edge；

(5) according to obtained outer edge center and radius, use butt joint ring for the constraint condition of concentric ring, inside and outside judgement The centre distance at edge is minimum, and the radius difference of outer edge is optimum with the comparable result of ring width is docked, final To the accurate central coordinate of circle of butt joint ring calculated from the binocular image comprising local butt joint ring target.

Above-mentioned steps (1) carry out binarization threshold segmentation using standard deviation method method particularly includes:

1) definition of standard deviation image is to calculate following standard deviation value to each pixel of image I:

Wherein, for any one pixel, u_k() be centered on the point around (2k+1) × (2k+1) neighborhood The average value of interior value, k indicate window half-breadth；

2) statistics with histogram is carried out to standard deviation image, binarization threshold ξ, judgment criteria deviation is calculated using averaging method Each pixel of image, pixel value are denoted as 255 more than or equal to ξ's, are denoted as 0 less than ξ, the bianry image after being divided.

Above-mentioned steps (1) carry out annulus region extraction using the area attribute in bianry image method particularly includes:

1) constraint condition set according to area, excessive for area or too small region are rejected；

2) in removing the image behind excessive region or region fine crushing, the solid degree of annulus region is smaller than other regions, By calculating solid degree, solid degree Minimum Area is judged for the annulus region that finally extracts.The calculating of solid degree is public Formula is as follows:

Carry out the specific method of fine and close reconstruction in above-mentioned steps (2) to the anchor ring extracted using half global matching process Are as follows:

Half global registration algorithm is along one-dimensional path r, cost value L of the pixel p to be matched under the conditions of parallax d_r(p,d) Recursive definition are as follows:

Wherein, min expression is minimized, and pixel indicates pixel p to be matched at parallax d matched data item C (p, d) Matching cost, matching cost selection sample insensitive method based on Birchfield and Tomasi and pixel intensity are calculated Difference.Section 2 indicates minimum cost value of adjacent pixel under the conditions of current pixel p parallax is d in path, and p-r indicates current road P's is upper under diameter.When the variation of a pixel unit occurs for the parallax of neighborhood pixels, cost value P₁, when parallax changes When more than a pixel unit, cost value P₂, P₁With P₂For given value.Last in above formula is chosen with current pixel p Parallax d it is unrelated, just to prevent L excessive, to meet L≤C_max+P₂Calculated result, C_maxIndicate the maximum of matching cost Value.I and k indicate the traversal of parallax from d_minTo d_max, d_minIndicate parallax minimum value, d_maxIndicate parallax maximum value.

Finally, the cost value of each pixel is all paths_rThe sum of cost value S (p, d)=∑_rL_r(p, d), it is each The matching result of point pixel is parallax corresponding to minimum cost value, so that the three-dimensional coordinate for completing each point in anchor ring calculates, It is i.e. fine and close to rebuild.

The three-dimensional reconstruction process under camera coordinates system is carried out in above-mentioned steps (2) are as follows: according to parallax under camera coordinates system As a result calculate the three-dimensional coordinate of each point, wherein camera coordinates system is defined as: cross camera photocentre, it is parallel as planar horizontal direction to The right side is x_cAxis, in parallel as plane vertical direction is downwards y_cAxis, z_cAxis meets the right-hand rule.

Multi-point fitting is used in above-mentioned steps (3), the method for largest optimization calculates anchor ring normal vector are as follows:

Three points are arbitrarily taken in the three dimensional point cloud of anchor ring, are carried out plane fitting, are obtained the normal vector of plane equation Indicate, then calculate other points arrive plane distance, distance and minimum when normal vector as final output, exactly anchor ring Normal vector.

Above-mentioned steps project to the three-dimensional point cloud indicated under camera coordinates system in the plane vertical with normal vector in (3) Are as follows:

The normal vector of plane is (p_h,q_h,r_h)(p_h,q_h,r_hRespectively represent the direction along tri- directions current coordinate system XYZ Number) and (p_h ²+q_h ²+r_h ²)=1, the coordinate of three-dimensional point cloud are (x_i,y_i,z_i) i=0,1 ... n-1, n are the sum of three-dimensional point cloud Mesh.Following coordinate transform then is carried out to three-dimensional point cloud:

Inward flange and outer edge are distinguished in above-mentioned steps (4) in the plane vertical with normal plane, using multi-point fitting, knot The optimal method of fruit carries out standard positive round fitting respectively, obtains center and the radius of butt joint ring outer edge, specifically:

The inward flange and outer edge obtained according to canny operator obtains inward flange and outer edge using the corresponding relationship of point The three-dimensional coordinate of point calculates the average distance between two groups of points of inward flange and outer edge；Then respectively in interior group of edge points or outer The fitting that three points carry out positive round center and radius is randomly selected in group of edge points, judges other points in the group and is fitted annulus Distance is considered optimum apart from summation is the smallest, exports respectively, obtain center and the radius of butt joint ring outer edge.

Compared with the prior art, the invention has the beneficial effects that: the present invention overcomes need docking in conventional method Hand labeled index point on ring and the shortcomings that consume a large amount of manpowers, while having avoided space annulus and having projected the throwing after two-dimensional surface The fitting problems for solving oval or conic section are not taken pains in shadow deformation, but by the measurement image of binocular camera acquisition, restore The three-dimensional information at its anchor ring and edge simultaneously projects it on the projection plane just penetrated, and the two dimension that is projected as simplifying space annulus is put down The positive round in face simplifies and calculates, it is only necessary to can accurately estimate anchor ring center according to the topography of annulus, guarantee certain robust Property under the premise of, have the advantages that estimation result is accurate, time-consuming low.

Detailed description of the invention

Fig. 1 is the step schematic diagram of method of the invention；

Fig. 2 (a)-(i) is embodiment 1 using the docking ring center extracting method extraction pair based on short distance short arc segments image Connect the treatment process of ring center；Wherein (a) is the short arc segments butt joint ring target left images pair of input, (b) after for polar curve correction Effect picture, (c) is standard deviation image, and (d) the binaryzation effect picture of standard deviation figure (e) is belonged to using the region of solid degree Property carry out annulus region extraction after effect picture, (f) effect picture of butt joint ring edge extracting, (g) dock ring region densification matching Obtained disparity map, (h) annular section Three-dimensional Gravity is laid foundations the effect picture of cloud re-projection, and (i) concentric circles fitting constraint obtains final The schematic three dimensional views in the center of circle.

Specific embodiment

Below with reference to Fig. 1, Fig. 2 (a)-(i), the present invention is described in detail.

(1) input image to be detected: image to be detected is the binocular gray level image comprising short arc segments butt joint ring, including left figure And right figure, as shown in Fig. 2 (a).

(2) correction of image polar curve is carried out to binocular image, as shown in Fig. 2 (b).Spin moment between known left and right camera Battle array is that (the corresponding Eulerian angles of R are A to R_lr), translation matrix be T when, polar curve correct the step of it is as follows:

1) in order to keep two camera coordinates systems parallel, first time rotation is carried out to two coordinate systems.Left camera rotates R_l=F (A_lr/ 2), right camera rotates R_r=F (- A_lr/ 2), F represents the transfer function between spin matrix and Eulerian angles herein, after rotation Translation vector T₂=R_rT；

2) in order to guarantee T₂Only in an X direction it is not 0 (i.e. base direction), second is carried out to Two coordinate system and is rotated, T is made₂It is flat Row Yu Xiangliangn=[- 10 0]^T。

(3) binocular image binary conversion treatment:

Wherein, for any one pixel, u_k() be centered on the point around (2k+1) × (2k+1) neighborhood The average value of interior value, k indicate window half-breadth, determine window size, are taken as 2 here.| | indicate signed magnitude arithmetic(al).Practical meter Calculation process using integrogram method promoted speed, calculate integrogram only need it is primary to image traversal, it is any using integrogram Pixel value in one rectangle and can be determined by four vertex, calculate unrelated with rectangular dimension, it is only necessary to which four sub-additions are transported It calculates.Shown in standard deviation figure such as Fig. 2 (c).

2) statistics with histogram is carried out to standard deviation image, the method for average calculates binarization threshold ξ, judgment criteria offset images Each pixel, pixel value is denoted as 255 more than or equal to ξ, less than ξ is denoted as 0.Shown in binary image such as Fig. 2 (d).

(4) annulus region is extracted using area attribute:

1) since there are regions fine crushing in binary picture non-circular region, size can effectively remove area fine crushing according to area Domain；

2) in removing the image behind region fine crushing, the solid degree of annulus region is smaller than other regions, can pass through meter Solid degree is calculated finally to extract annulus region:

The extraction effect of annulus region is as shown in (e) in Fig. 2.

(5) edge extracting that annulus region is carried out using canny operator, obtains outer edge point position.Extraction effect is such as Shown in Fig. 2 (f).

(6) it because the texture of anchor ring is less, commonly uses matching process and is easy failure, anchor ring cause is carried out by half global method Close reconstruction.

Shown in disparity map effect such as Fig. 2 (g) that densification is rebuild, by taking the anchor ring extracted in left image as an example.

(7) solve plane of a loop equation according to the anchor ring of reconstruction: the present invention randomly chooses on anchor ring obtained in (6) step Any 3 three-dimensional point X_i,Y_i,Z_iI=1,2,3 calculate the parameter of one group of plane equation, and plane equation is as follows:

AX+bY+cZ+d=0 (4)

Wherein, a, b, c, d are the expression coefficient of plane equation.General plane equation is without going past coordinate axis center (anchor ring Have certain vertical range from camera), therefore above formula can simplify are as follows:

AX+BY+CZ=-1 (5)

By X_i,Y_i,Z_iI=1,2,3 bring above formula into, solve A, B, C.Then other three-dimensional point (x are calculated_j,y_j,z_j) arrive The distance of current planeAll distances are cumulative, and distance and the smallest result are exported, A, B, C are obtained, plane normal vector (p is converted to_h,q_h,r_h), output In order to improve computational efficiency and obtain convergent as a result, by multiple Experiment statistics can obtain the smallest one group of result of ideal distance generally after 200 random selection calculating parameters.

(8) three-dimensional point cloud is projected in the plane vertical with normal vector according to the plane equation in step (7), is specifically done Method is:

The normal vector of plane is (p_h,q_h,r_h) and (p_h ²+q_h ²+r_h ²)=1, the coordinate of three-dimensional point cloud are (x_i,y_i,z_i) i= 0,1 ... n-1, n are the total number of three-dimensional point cloud.Following coordinate transform then is carried out to three-dimensional point cloud, obtains new three-dimensional coordinate (x'_i,y'_i,z'_i):

Shown in the three-dimensional point cloud effect such as Fig. 2 (h) projected in normal vector vertical plane.

(9) positive round fitting is carried out to the coordinate after projection, specific practice is:

Equation of a circle is (x'_k-x₀)²+(y'_k-y₀)²=R₀ ², wherein (x'_k,y'_k) it is to be chosen in the obtained point cloud of (8) step Inner annular edge point, k=1,2,3 three points can solve, x₀、y₀And R₀For amount to be asked.X is acquired using least square method₀、y₀With R₀.Then by other inner annular edge point (x'_l,y'_l,z'_l) substitute into ∑ ((x'_l-x₀)²+(y'_l-y₀)²-R₀ ²), overall error minimum is then Export x₀、y₀And R₀。

Similarly, (x' is taken_m,y'_m) it is the outer ring marginal point chosen in point cloud that (8) step obtains, m=1,2,3 three points are It can solve, x₁、y₁And R₁For amount to be asked.X is acquired using least square method₁、y₁And R₁.Then by other outer ring marginal point (x'_n, y'_n,z'_n) substitute into ∑ ((x'_n-x₁)²+(y'_n-y₁)²-R₁ ²), overall error minimum then exports x₁、y₁And R₁。

(10) evaluation output is carried out to last result

Two center of circle x that (9) step obtains₀、y₀、x₁、y₁With radius R₀, R₁, judge the distance and radius of center location Then (x, y) is rotated back to before normal vector projection by difference less than the x and y coordinates as the output center of circle of given threshold Plane obtains the Spatial outlier value (x in the center of circle_center,y_center,z_center).Determine butt joint ring center and the anchor ring three-dimensional effect of output Shown in fruit such as Fig. 2 (i).

Experimental example:

To (a) in image to be detected Fig. 2, successively handled according to each step in the present invention, from result The present invention can accurately detect circle ring area, and can be effectively carried out the reconstruction of plane of a loop and the estimation in the center of circle.

Claims

1. a method for extracting the center of a docking ring based on a short-distance short-arc segment image, is characterized in that the realization steps are as follows:

(1) Perform epipolar correction for the binocular image containing the short-arc docking ring target obtained at close range, and then use the standard deviation method to perform binarization threshold segmentation to obtain the segmented binary image. Extract the torus region based on the regional attributes of the torus, and use the canny operator to extract its inner and outer edges after the torus region extraction;

(2) Use the semi-global matching method to perform dense reconstruction on the extracted torus area, and obtain the parallax value of a large number of points on the torus, so as to perform 3D reconstruction in the camera coordinate system, and obtain a 3D point cloud; the large number of points is all the pixels that can be registered on the torus, which is determined by the size of the imaging area and the registration rate of the torus;

(3) Multi-point fitting is used for the 3D point cloud in the camera coordinate system, and the maximum optimization method is used to calculate the normal vector of the torus and project the 3D point cloud represented in the camera coordinate system into a plane perpendicular to the normal vector;

(4) Distinguish the inner edge and the outer edge in a plane perpendicular to the normal plane, and use multi-point fitting. The method with the best result is to perform standard perfect circle fitting respectively to obtain the center and radius of the inner and outer edges of the butt ring;

(5) If the difference between the center position and the radius of the inner and outer edges of the butt ring obtained in step (4) is smaller than the set threshold, then according to the obtained center and radius of the inner and outer edges, the constraint condition that the butt ring is a concentric ring is adopted, and finally the The exact circle center coordinates of the docking ring calculated in the binocular image containing the local short arc segment docking ring target.

2. a kind of method for extracting the center of a docking ring based on a short-distance short arc segment image according to claim 1, is characterized in that: in the described step (1), the concrete method that adopts standard deviation method to carry out binarization threshold segmentation is as follows: :

(11) The definition of the standard deviation image is that the following standard deviation value is calculated for each pixel of the image I:

Among them, for any pixel point, u _k ( ) is the average value of the values in the (2k+1)×(2k+1) neighborhood around the point, and k represents the half-width of the window;

(12) Perform histogram statistics on the standard deviation image, use the mean method to calculate the binarization threshold ξ, and judge each pixel of the standard deviation image. The pixel value greater than or equal to ξ is recorded as 255, and the pixel value less than ξ is recorded as 0, to obtain The segmented binary image.

3 . The method for extracting the center of a butt ring based on a short-range short-arc segment image according to claim 1 , wherein the step (1) utilizes the region attribute in the binary image to extract the specific torus region. 4 . The method is:

(21) According to the constraints set by the area, the areas that are too large or too small are eliminated;

(22) In the image after removing the excessively large area or the finely divided area, the solid degree of the torus area is smaller than that of other areas. By calculating the solid degree, it is judged that the area with the smallest solid degree is the final extracted torus area. Calculation of the solid degree The formula is as follows:

4 . The method for extracting the center of a butt ring based on a short-range short-arc segment image according to claim 1 , wherein the step (2) utilizes a semi-global matching method to perform dense reconstruction on the extracted torus. 5 . The specific method is:

The semi-global matching algorithm follows a one-dimensional path r, and the cost value L _r (p, d) of the pixel p to be matched under the condition of disparity d is recursively defined as:

Among them, min represents the minimum value, and the pixel pair matching data item C(p, d) represents the matching cost of the pixel p to be matched under the parallax d. The matching cost is calculated by the method based on Birchfield and Tomasi sampling insensitive to calculate the pixel brightness difference, The second item represents the minimum cost value of adjacent pixels in the path under the condition that the disparity of the current pixel p is d, and pr represents the previous point of p in the current path. When the disparity of adjacent pixels changes by one pixel unit, the cost value is P _1. When the parallax change exceeds one pixel unit, the cost value is P ₂ , and P ₁ and P ₂ are given values; the last item in the above formula has nothing to do with the parallax d selected by the current pixel p, only to prevent L from being too large. Large, to satisfy the calculation result of L≤C _max +P ₂ , C _max represents the maximum value of the matching cost, i and k represent the traversal of the parallax from d _min to d _max , d _min represents the minimum value of the parallax, and d _max represents the parallax maximum value;

Finally, the cost value of each pixel is the sum of the cost values of all paths r S(p,d)=∑ _r L _r (p,d), the matching result of each pixel is the disparity corresponding to the minimum cost value, Thus, the three-dimensional coordinate calculation of each point in the torus is completed, that is, the dense reconstruction.

5. The method for extracting the center of a docking ring based on a short-distance short-arc segment image according to claim 1, wherein: in the step (2), the three-dimensional reconstruction process under the camera coordinate system is: Under the system, the three-dimensional coordinates of each point are calculated according to the parallax results, where the camera coordinate system is defined as: passing through the camera optical center, the horizontal direction parallel to the image plane is the x _c axis to the right, and the vertical direction parallel to the image plane downward is the y _c axis. , the z- _c axis satisfies the right-hand rule.

6. The method for extracting the center of a docking ring based on a short-distance short-arc segment image according to claim 1, wherein: in the step (3), multi-point fitting is adopted, and the method of maximum optimization calculates the torus method The vector is:

Take three points arbitrarily in the 3D point cloud data of the torus, perform plane fitting, get the normal vector representation of the plane equation, and then calculate the distance from other points to the plane, the distance and the minimum normal vector are used as the final output, which is The normal vector of the torus.

7. The method for extracting the center of a docking ring based on a short-distance short-arc segment image according to claim 1, wherein: in the step (3), the three-dimensional point cloud represented by the camera coordinate system is projected onto the and method The vector in the vertical plane is:

The normal vector of the plane is (ph , q _h , _rh ), _ph , q _h , _rh represent the direction numbers along the three directions of the current coordinate system XYZ respectively ₎ and _ph ² +q _h ² +r _h ² =1, the coordinates of the 3D point cloud are (x _i , y _i , z _i ), i=0,1,...n-1, where n is the total number of 3D point clouds, then the following coordinate transformation is performed on the 3D point cloud:

8. The method for extracting the center of a butt ring based on a short-range short-arc segment image according to claim 1, wherein in the step (4), the inner edge and the outer edge are distinguished in a plane perpendicular to the normal plane , using multi-point fitting, and the method with the best results performs standard perfect circle fitting respectively, and obtains the center and radius of the inner and outer edges of the butt ring, specifically:

According to the inner edge and outer edge obtained by the canny operator, the three-dimensional coordinates of the inner edge and the outer edge point are obtained by using the corresponding relationship of the points, and the average distance between the two groups of points on the inner edge and the outer edge is calculated; Or randomly select three points in the outer edge point group to fit the center and radius of the perfect circle, judge the distance between other points in the group and the fitted circle, the smallest sum of distances is considered as the best result, and output them respectively to get the docking ring The center and radius of the inner and outer edges.