CN104036516A - Camera calibration checkerboard image corner detection method based on symmetry analysis - Google Patents

Abstract

The invention discloses a camera calibration checkerboard image corner detection method based on symmetry analysis. The method comprises the steps that firstly, an input road image is preprocessed; then, candidate feature points are extracted from a checkerboard image according to the characteristic that checkerboard corners have the structural symmetry, and image feature points are detected through candidate feature point clusters and ChESS operators; finally, according to the characteristic that the corners of the checkerboard image exist in the image feature points and have the structural symmetry, detection of all the corners of the checkerboard image is achieved by calculating the initial corner first and then calculating the positions of the other corners. According to the method, the symmetry of the checkerboard image is comprehensively considered, and automatic detection of the corners of the checkerboard image is achieved. The method has great significance for automatic camera calibration. Due to the fact that background feature points and the corners of the checkerboard image can be effectively distinguished, the method has high robustness.

Description

Camera calibration checkerboard image angular-point detection method based on symmetrical analysis

Technical field

The present invention relates to image and process and field of machine vision, especially image is processed and a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis of field of machine vision.

Background technology

In image measurement and field of machine vision, camera calibration refers to the process of asking for camera interior and exterior parameter according to camera model, is from two dimensional image, to obtain the basic step of three-dimensional information, is widely used in the fields such as three-dimensional reconstruction, vision-based detection and monitoring.In existing camera marking method, the plane template scaling method based on black and white checkerboard image is widely used because calibrating template is made simply, scaling method is ripe.Because camera calibration precision depends on the positioning precision of calibrating template angle point to a great extent, the corner location that therefore how to detect calibrating template is the important subject of field of machine vision.

In camera calibration experiment, because checkerboard image often has more complicated background, for reducing the impact of background image on checkerboard image Corner Detection, the man-machine interaction method that mostly adopts mouse to click at present carries out Corner Detection.First the method need to adopt mouse to click in certain sequence 4 points of checkerboard image periphery, in order to make the term of reference of corner location, and then detects the angle point within the scope of click by Corner Detection Algorithm.This method wastes time and energy, and is difficult to realize the robotization of camera calibration process.Other researchs pay close attention to how to improve conventional angular point detecting method to solve checkerboard image Corner Detection problem, and to extract Study on Problems less to having the angle point of complex background checkerboard image.

At present, some checkerboard image angle point extraction methods have been proposed.As adopt hough conversion to detect checkerboard image angle point by extracting image linear feature, to realize the automatic extraction of checkerboard image unique point.Generally speaking, compare with camera marking method, the method that automatically detects and extract for checkerboard image angle point specially proposing is at present still less, and this has hindered the raising of camera calibration process automation degree to a certain extent.

Summary of the invention

Object of the present invention is exactly in order to address the above problem, a kind of new camera calibration checkerboard image angular-point detection method based on symmetrical analysis is provided, the method considers the symmetry that checkerboard image has, and has realized the automatic detection of checkerboard image angle point, is applicable to camera calibration.

To achieve these goals, the present invention adopts following technical scheme:

A camera calibration checkerboard image angular-point detection method based on symmetrical analysis, comprising: carry out checkerboard image and read; The image reading is carried out to rim detection; Edge image carries out binaryzation; On the edge image of binaryzation, carry out candidate feature point and detect, obtain candidate feature point; The candidate feature point detecting is carried out to cluster; All candidate feature points in each candidate feature point cluster are carried out to symmetry calculating, obtain unique point set; In unique point set, carry out initial Corner Detection; Take initial angle point as initial search point, estimate other corner location, in the error range of position, the unique point at estimated corner location place is set to angle point; To the angle point searching is capable, judge; Export final angle point.

Concrete steps are as follows:

Step (1): start, read checkerboard image;

Step (2): adopt canny operator to carry out rim detection to image, obtain edge image;

Step (3): edge image carries out binaryzation, obtains the edge image of binaryzation;

Step (4): carry out candidate feature point and detect on the edge image of binaryzation, by the moving window traversal binaryzation edge image that size is w * w is set, if there are 4 intersection points at the edge in the edge of moving window and binaryzation edge image, the central point of moving window are set to candidate feature point, thereby obtain the set of candidate feature point;

Step (5): adopt the method for picture point mark that the candidate feature point detecting is carried out to cluster, obtain the cluster of candidate feature point;

Step (6): adopt ChESS operator to calculate the symmetrical response of all candidate feature points in each candidate feature point cluster;

Step (7): judge according to the symmetrical response of candidate feature point whether this point is unique point, if it is this point is put into unique point set, if not casting out this point;

Step (8): whether the extraction that judges all candidate feature point clusters institute character pair point completes, if completed, enters step (9); If do not completed, return to step (6);

Step (9): in unique point set, carry out initial Corner Detection;

Step (10): take initial angle point as initial search point, estimate other corner location, in the error range of position, the unique point at corner location place is set to angle point; The angle point searching is arranged by row, and claimed that this row angle point is that angle point is capable; The number of contained angle point in checkerboard image a line angle point is set, capable if the angle point number containing in a line angle point searching and identical this row angle point of angle point number of setting are set to normal angle point, if angle point number difference is set to improper angle point capable;

Step (11): whether contain two adjacent normal angle points during the angle point that searched of judgement is capable capable; If it is capable two adjacent normal angle points to be detected, adopt normal angle point capable of the capable alignment of data that carries out of the angle point having searched, make that angle point is capable meets symmetry requirement, enter step (12); If search finishes not find two adjacent normal angle points capable, current checkerboard image is processed as abnormal conditions, enter step (13);

Step (12): output image angle point,

Step (13) finishes.

The candidate feature point that the method that adopts picture point to mark in described step (5) obtains detection carries out cluster, and the method that obtains the set of candidate feature point cluster is:

First candidate feature point is marked, compose the identical label of neighboring candidate unique point, will there is the connected domain of the candidate feature point of same numeral as candidate feature point; Minimum connected domain distance is set, and the different connected domains that distance are less than to this setpoint distance are merged into a connected domain, and regard the connected domain after each merging a cluster of candidate feature point as, thereby realize the cluster of candidate feature point.

In described step (6), adopt the method for candidate feature point symmetry response in ChESS operator calculated candidate feature points clustering to be:

If candidate feature point set P _fdbe divided into N candidate feature point cluster { s ₁, s ₂..., s _n, cluster s _iin contain m candidate feature point { d ₁, d ₂..., d _m, calculate s _iin the symmetry analog value of each candidate feature point, by s _imiddle symmetrical response is the highest and be greater than 0 candidate feature point and put into unique point set as unique point.

The formula of the symmetrical response of described calculated candidate unique point is

$R\left(f_i\right)=\underset{j=1}{\overset{m}{\max }}\left(R\left(d_j\right)\right)---\left(1\right)$

Wherein, R (d _j) expression candidate feature point d _jsymmetrical response; R(d _j) computing formula be:

R(d _j)＝SR-DR-MR (2)

Wherein, SR, DR and MR represent respectively overall response value, differential response value and the average response of candidate feature point.Suppose with d _jfor sampling center, it is r that sample radius is set, and sampled point number is n, chooses sampling point set I={I in checkerboard image ₁, I ₂..., I _n, the computing formula of SR, DR and MR is respectively:

$\mathrm{SR}=\underset{k=1}{\overset{n/4}{\mathrm{\Σ}}}|(I_k+I_{k+n/2})-(I_{k+1/4}+I_{k+3n/4})|---\left(3\right)$

$\mathrm{DR}=\underset{k=1}{\overset{n/2}{\mathrm{\Σ}}}|I_k-I_{k+n/2}|---\left(4\right)$

$\mathrm{MR}=|\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{I}_k-\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{I}_k^{\′}|---\left(5\right)$

Wherein, while calculating MR, represent d _jadjacent image point in checkerboard image, gets d _j8 Neighborhood Graph picture point conducts

The method of carrying out initial Corner Detection in described step (9) is:

Feature points set P _fin with unique point p ₀centered by comprise p ₀in interior any 9 adjacent feature points, distance center unique point p ₀4 nearest unique point p ₁～p ₄for closing on unique point, central feature point p ₀the unique point p of diagonal ₅～p ₈for diagonal angle unique point;

If above-mentioned unique point set P _f9 unique points of middle arbitrary neighborhood meet closes on unique point symmetry, diagonal angle unique point symmetry and closes on the requirement of unique point angular range, these 9 adjacent unique points is defined as to initial angle point.

Described close on unique point symmetry require be:

Suppose unique point p ₀～p ₈coordinate be respectively (x _i, y _i), 0≤i≤8, if in the error range of position, close on unique point p ₁～p ₄coordinate and p ₀between coordinate, meet following relation:

$\left\{\begin{array}{c}{x}_0=\frac{x_j+x_{j+2}}{2}\\ y_0=\frac{y_j+y_{j+2}}{2}\end{array}\right.---\left(6\right)$

Claim to close on unique point p ₁～p ₄about central feature point p ₀symmetry, meets and closes on the requirement of unique point symmetry; Wherein, (x ₀, y ₀) centered by unique point p ₀coordinate, (x _j, y _j) represent to close on the coordinate of unique point, j=1,2.

Described diagonal angle unique point symmetry requires:

Suppose that diagonal angle unique point is p _k, 5≤k≤8, the unique point of closing on that it is adjacent is respectively p _iand p _j, 1≤i, j≤4, if in the error range of position p _kcoordinate and p _iand p _jand p ₀between coordinate, meet:

$\left\{\begin{array}{c}{x}_c=\frac{x_i+x_j}{2}\\ y_c=\frac{y_i+y_j}{2}\\ x_k={2x}_c-x_0\\ y_k={2y}_c-y_0\end{array}\right.---\left(9\right)$

Claim diagonal angle unique point p ₅～p ₈meet symmetry requirement; (x _i, y _i), (x _j, y _j), (x ₀, y ₀), (x _k, y _k) represent respectively p _i, p _j, p ₀and p _kcoordinate, (x _c, y _c) for closing on unique point p _iand p _jthe coordinate of central point.

Described close on unique point angular range require be:

Close on unique point p ₁～p ₄with p ₀between angle can be expressed as:

${\mathrm{\θ}}_{p_i{p}_0{p}_{i+1}}=\arccos \left(\frac{d_1^2+d_2^2-d_3^2}{{2d}_1{d}_2}\right)---\left(10\right)$

Wherein, d ₁, d ₂and d ₃represent respectively p _iwith p ₀, p _i+1with p ₀, p _iwith p _i+1between distance,

$d_1=\sqrt{{(x_i-x_0)}^2+{(y_i-y_0)}^2},$ $d_2=\sqrt{{(x_{i+1}-x_0)}^2+{(y_{i+1}-y_0)}^2},$ $d_3=\sqrt{{(x_{i+1}-x_i)}^2+{(y_{i+1}-y_i)}^2},$ (x _i, y _i), (x _i+1, y _i+1), (x ₀, y ₀) represent respectively p _i, p _i+1and p ₀coordinate, when i=4, get p _i+1=p ₁;

If meet:

${\mathrm{\θ}}_{p_i{p}_0{p}_{i+1}}\≥{\mathrm{\θ}}_{\min }---\left(11\right)$

Claim to close on unique point and central feature point meets angle requirement; Wherein, θ _minfor the minimum angle value arranging, get θ _min>=15 °.

The concrete grammar of described step (10) is:

By other corner location of known angle point estimation, comprise following three kinds of situations:

1) if above angle points continuously of 2 and 2 in known a line are estimated other angle point in this row: suppose p ₁and p ₂represent respectively two adjacent corner points of angle point in capable, calculate respectively p ₁about p ₂symmetric points p ₃and p ₂about p ₁symmetric points p ₄position, within the scope of position estimation error, extract P _fmiddle p ₃and p ₄the unique point of position is as this row angle point; Adopt successively the method, realize this angle point capable in the detection of all angle points.

2) capable if the angle point having detected comprises two adjacent normal angle points in capable, according to normal angle point capable estimate other angle point capable in the position of angle point: suppose p ₁₁, p ₁₂and p ₁₃for the angle point in normal angle point capable 1, p ₂₁, p ₂₂and p ₂₃angle point in normal angle point capable 2, calculates respectively p ₁₁about p ₂₁symmetric points p ₃₁, p ₂₁about p ₁₁symmetric points p ₀₁, p ₁₂about p ₂₂symmetric points p ₃₂, p ₂₂about p ₁₂symmetric points p ₀₂, p ₁₃about p ₂₃symmetric points p ₃₃, p ₂₃about p ₁₃symmetric points p ₀₃; P ₀₁, p ₀₂, p ₀₃for the position of angle point in adjacent corner points row 0, p ₃₁, p ₃₂, p ₃₃position for angle point in adjacent corner points row 3; In the error range of position, by unique point set P _fthe unique point of middle corresponding position is set to angle point.

3) if the angle point having detected does not have two adjacent normal angle points capable in capable, according to existing angle point capable estimate other angle points capable in the position of angle point: suppose p ₁, p ₂and p ₃for in the capable H of angle point ₀three adjacent angle points in centre position, H ₁for H ₀adjacent corner points row, p ₄for in H ₁the angle point in centre position, calculates respectively p ₁, p ₂and p ₃about p ₄symmetric points p ₅, p ₆and p ₇, p ₅～p ₇for the new capable H of angle point ₂in corner location, in the error range of position, by P _fat p ₅～p ₇the unique point of position is set to angle point.

Beneficial effect of the present invention: the present invention proposes a kind of checkerboard image angle point automatic testing method based on symmetrical analysis.The method has symmetric feature according to checkerboard image, by checkerboard image candidate feature point is carried out to cluster, adopts ChESS operator to calculate the corresponding unique point of each cluster, realizes the extraction of checkerboard image unique point; Adopt and first to detect initial angle point and then take initial angle point and detect the method for other angle point as starting point, realize the Corner Detection of checkerboard image.The checkerboard image angle point that the method that the present invention proposes detects has good precision, for solving checkerboard image angle point, automatically detects new thinking is provided, and to realizing video camera automatic Calibration, has positive effect.

Accompanying drawing explanation

Fig. 1 is the camera calibration checkerboard image angular-point detection method scheme process flow diagram that the present invention is based on symmetrical analysis;

Fig. 2 is the initial angle point schematic diagram of the present invention;

Fig. 3 is checkerboard image sampled point schematic diagram;

Fig. 4 (a) is 2 and 2 above angle points continuously in the known a line of the present invention, estimates the schematic diagram of this other angle point of row;

Fig. 4 (b) is for the present invention is according to 2 capable schematic diagram of estimating that other angle point is capable of adjacent improper angle point;

Fig. 4 (c) is for the present invention is according to known 2 capable schematic diagram of estimating that other angle point is capable of adjacent normal angle point.

Embodiment

Below in conjunction with accompanying drawing and embodiment, the invention will be further described.

Realizing the required basic hardware condition of system architecture of the present invention is: a dominant frequency is 2.4GHZ, inside saves as the computing machine of 1G, and required software condition is: programmed environment is Visual C++6.0.

As shown in Figure 1, the camera calibration checkerboard image angular-point detection method based on symmetrical analysis, concrete steps are as follows:

Step (1): start, read checkerboard image;

Step (2): adopt canny operator to carry out rim detection to image, obtain edge image;

Step (3): edge image carries out binaryzation, obtains the edge image of binaryzation;

Step (4): carry out candidate feature point and detect on the edge image of binaryzation, by the moving window traversal binaryzation edge image that size is w * w is set, if there are 4 intersection points at the edge in the edge of moving window and binaryzation edge image, the central point of this moving window is set to candidate feature point, obtain the set of candidate feature point.

Step (5): adopt the method for picture point mark to carry out cluster to candidate feature point, obtain candidate feature point cluster;

First candidate feature point is marked, compose the identical label of neighboring candidate unique point, will there is the connected domain of the candidate feature point of same numeral as candidate feature point; Minimum connected domain distance is set, and the different connected domains that distance are less than to this setpoint distance are merged into a connected domain, and regard the connected domain after each merging a cluster of candidate feature point as, thereby realize the cluster of candidate feature point.Concrete steps are:

(step 5-1): adopt labeling algorithm to carry out mark to image candidate feature point, obtain marking image ImBj;

(step 5-2): press from top to bottom, order is from left to right got respectively the border of each connected region in ImBj, search radius searchR is set, whether calculating there is other adjacent connected region apart from each connected region border within the scope of searchR, if existed, this adjacent connected region is merged, otherwise connected region is constant.

Step (6): adopt ChESS operator to carry out symmetrical response calculating to candidate feature points all in each candidate feature point cluster; Method is: by image labeling and connected domain, merge candidate feature point set P _fdbe divided into N candidate feature point cluster { s ₁, s ₂..., s _n, suppose cluster s _iin contain m candidate feature point { d ₁, d ₂..., d _m, extract minutiae set P _fmain thought be by calculating s _iin the symmetry of each candidate feature point, select s _iin there is the candidate feature point of high symmetrical response and put into unique point set as unique point, the formula that calculates symmetrical response is

$R\left(f_i\right)=\underset{j=1}{\overset{m}{\max }}\left(R\left(d_j\right)\right)---\left(1\right)$

Wherein, R (d _j) expression candidate feature point d _jsymmetrical response.

The concrete steps of step (6) are:

Step (6-1) is supposed cluster s _iin contain M picture point, it is r that sample radius is set, sampled point number is n, with s _iin each picture point d _jcentered by, in checkerboard image Im, choose sampling point set I={I ₁, I ₂..., I _n, its sampling process is as shown in Figure 3.In figure, provide sample radius and be respectively r=3 and r=5, during number of samples n=16, the relation between sampling center 0 and sampled point.Sampled point number n gets 4 integral multiple.

Step (6-2) is according to picture point p _jsampling point set I obtain d _jsymmetrical response be:

R＝SR-DR-MR (2)

Wherein, SR, DR and MR are respectively picture point d _joverall response value, differential response value and average response, its calculating formula can be expressed as:

$\mathrm{SR}=\underset{k=1}{\overset{n/4}{\mathrm{\Σ}}}|(I_k+I_{k+n/2})-(I_{k+1/4}+I_{k+3n/4})|---\left(3\right)$

$\mathrm{DR}=\underset{k=1}{\overset{n/2}{\mathrm{\Σ}}}|I_k-I_{k+n/2}|---\left(4\right)$

$\mathrm{MR}=|\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{I}_k-\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{I}_k^{\′}|---\left(5\right)$

Wherein, while calculating MR, for d in checkerboard image Im _jnear the picture point of 8 neighborhoods.

Step (6-3) employing formula (1) is calculated cluster s _iin the symmetrical response of all picture point, if certain point has maximum symmetrical response and response is greater than 0, using this picture point as s _icorresponding image characteristic point.

Step (6-4) is extracted respectively candidate feature point set P according to said method _fdmiddle N candidate feature point cluster { s ₁, s ₂..., s _ncorresponding unique point, feature point set obtained.

Step (7): judge according to the symmetrical response of candidate feature point whether this point is unique point, if it is this point is put into unique point set, if not casting out this point;

Step (8): judgement P _fdwhether all candidate feature point clusters have been calculated, if completed, enter step (9); If do not completed, return to step (6);

Step (9): in feature point set resultant vector, carry out initial Corner Detection;

The method of carrying out initial Corner Detection is: whether 9 unique points putting arbitrary neighborhood in set by judging characteristic meet and close on unique point symmetry, diagonal angle unique point symmetry and close on the requirement of unique point angular range, if met, are defined as initial angle point.

Defining initial angle point is unique point set P _fin meet arbitrarily 9 adjacent feature points that symmetry requires, its relation is as shown in Figure 2.In Fig. 2, central feature point is p ₀, adjacent feature point is respectively p ₀～p ₈.Wherein, p ₁～p ₄for apart from p ₀4 nearest unique points, are called and close on unique point; p ₅～p ₈for p ₀the unique point of diagonal, is called diagonal angle unique point.Unique point set P _fin 9 adjacent unique points whether be that initial angle point need to judge from three aspects:, close on unique point symmetry, diagonal angle unique point symmetry and close on unique point angular range.

(9-1) closing on unique point symmetry requires computing method to be

Suppose unique point p ₀～p ₈coordinate is respectively (x _i, y _i), 0≤i≤8, if close on unique point p in the error range of position ₁～p ₄with p ₀coordinate meets following relation:

$\left\{\begin{array}{c}{x}_0=\frac{x_j+x_{j+2}}{2}\\ y_0=\frac{y_j+y_{j+2}}{2}\end{array}\right.---\left(5\right)$

Claim to close on unique point p ₁～p ₄about central feature point p ₀symmetry, meets and closes on the requirement of unique point symmetry; Wherein, (x ₀, y ₀) centered by unique point p ₀coordinate, (x _j, y _j) represent to close on the coordinate of unique point, j=1,2.

(9-2) closing on unique point angular range calculates

According to closing on unique point and central feature point coordinate, p ₁～p ₄with p ₀between angle can be expressed as:

${\mathrm{\θ}}_{p_i{p}_0{p}_{i+1}}=\arccos \left(\frac{d_1^2+d_2^2-d_3^2}{{2d}_1{d}_2}\right)---\left(6\right)$

Wherein, d ₁, d ₂and d ₃represent respectively p _iwith p ₀, p _i+1with p ₀, p _iwith p _i+1between distance, $d_1=\sqrt{{(x_i-x_0)}^2+{(y_i-y_0)}^2},$ $d_2=\sqrt{{(x_{i+1}-x_0)}^2+{(y_{i+1}-y_0)}^2},$ $d_3=\sqrt{{(x_{i+1}-x_i)}^2+{(y_{i+1}-y_i)}^2},$ (x _i, y _i), (x _i+1, y _i+1), (x ₀, y ₀) represent respectively p _i, p _i+1and p ₀coordinate, when i=4, get p _i+1=p ₁.If meet:

${\mathrm{\θ}}_{p_i{p}_0{p}_{i+1}}\≥{\mathrm{\θ}}_{\min }---\left(7\right)$

Claim to close on unique point and central feature point meets angle requirement.Wherein, θ _minfor the minimum angle value arranging, get θ _min>=15 °.

(9-3) diagonal angle unique point symmetry computing method are:

Suppose that diagonal angle unique point is p _k, 5≤k≤8, the unique point of closing on that it is adjacent is respectively p _iand p _j, 1≤i, j≤4, if p in the error range of position _kcoordinate and p _iand p _jand p ₀coordinate meet:

$\left\{\begin{array}{c}{x}_c=\frac{x_i+x_j}{2}\\ y_c=\frac{y_i+y_j}{2}\\ x_k={2x}_c-x_0\\ y_k={2y}_c-y_0\end{array}\right.---\left(8\right)$

Claim diagonal angle unique point p ₅～p ₈meet symmetry requirement.(x _i, y _i), (x _j, y _j), (x ₀, y ₀), (x _k, y _k) represent respectively p _i, p _j, p ₀and p _kcoordinate, (x _c, y _c) for closing on unique point p _iand p _jthe coordinate of central point.

Step (10): take initial angle point as initial search point, by estimating other corner location, in the error range of position, the unique point of relevant position in unique point set is set to angle point;

The angle point searching is arranged by row, and claimed that this row angle point is that angle point is capable; The number of contained angle point in checkerboard image a line angle point is set, capable if the angle point number containing in a line angle point searching and identical this row angle point of angle point number of setting are set to normal angle point, if angle point number difference is set to improper angle point capable.During angle point is capable, putting in order of angle point puts in order identical with the angle point in camera calibration template.Define the angle point that in a line angle point that normal angle point behavior detects, angle point number is preset value capable.This preset value is the angle point number that in camera calibration template, a line angle point comprises.The method that unique point in unique point set is set to angle point in the error range of position comprises following three kinds of situations: the continuous angle point in (one) known a line more than 2 and 2, estimate the position of other angle point in this row; (2) according to 2 adjacent improper angle points are capable, estimate the angle point of other angle point in capable; (3) known 2 adjacent normal angle points are capable estimates the angle point of other angle point in capable.

Described (one) circular is: the corner location method of estimation of employing is as shown in Fig. 4 (a).Wherein, p ₁and p ₂represent respectively two adjacent corner points of angle point in capable.According to symmetry, can estimate respectively p ₁about p ₂symmetric points p ₃and p ₂about p ₁symmetric points p ₄position.Within the scope of position estimation error, extract P _fmiddle p ₃and p ₄the unique point of position is as this row angle point.Adopt successively the method, realize this angle point capable in the detection of all angle points.

Described (two) circular is: the corner location method of estimation of employing is as shown in Fig. 4 (b).Wherein, p ₁, p ₂and p ₃for three adjacent corner points of certain angle point in capable, p ₄and p ₀be respectively the angle point in adjacent corner points row, p ₅～p ₁₀for the new angle point that estimates according to symmetry capable in the position of angle point.In the error range of position, by p ₅～p ₁₀the unique point of position is set to angle point, then respectively with p ₅～p ₇and p ₈～p ₁₀for starting point, adopt method shown in Fig. 4 (a) to detect the angle point that makes new advances other angle point in capable.

Described (three) circular is: the corner location method of estimation of employing is as shown in Fig. 4 (c).Suppose p ₁₁, p ₁₂and p ₁₃for the angle point in normal angle point capable 1, p ₂₁, p ₂₂and p ₂₃for the angle point in normal angle point capable 2, according to symmetry, can estimate respectively the position of angle point in adjacent corner points row 0 and angle point capable 3.In the error range of position, by unique point set P _fthe unique point of middle relevant position is set to angle point.

In said process, according to two adjacent corner points, estimate that the method for other corner location is: suppose any two adjacent corner points p _iand p _jcoordinate be respectively (x _i, y _i), (x _j, y _j), p _iabout p _jsymmetrical angle point p _si=(x _si, y _si) and p _jabout p _isymmetrical angle point p _sj=(x _sj, y _sj) can be expressed as:

$\left\{\begin{array}{c}{x}_{\mathrm{sj}}={2x}_i-x_j\\ y_{\mathrm{sj}}={2y}_i-y_j\end{array}\right.---\left(9\right)$

$\left\{\begin{array}{c}{x}_{\mathrm{si}}={2x}_j-x_i\\ y_{\mathrm{si}}={2y}_j-y_i\end{array}\right.---\left(10\right)$

The concrete operation step of realizing above-mentioned (one), (two) and (three) is:

Step (10-1): according to initial angle point adopt method shown in Fig. 4 (a) estimate initial angle point place angle point capable in the position of other angle point, in the error range of position, search for individual features point as angle point.

Step (10-2): comprise two adjacent normal angle points in as capable in the angle point having detected capable, adopt method shown in Fig. 4 (c) to estimate the position of the capable angle point of other angle point; If it is capable two adjacent normal angle points not detected, according to method shown in existing angle point capable employing Fig. 4 (b), estimate the corner location that other angle point is capable.

Step (10-3): the capable minimum angle point number of angle point is set, if the angle point detecting capable in the number of angle point be less than the minimum angle point number of setting, changing angle point line search direction, again to search for angle point capable.

Step (11): whether contain two adjacent normal angle points during the angle point that searched of judgement is capable capable.If it is capable to contain two adjacent normal angle points, adopt normal angle point capable of the capable alignment of data that carries out of the angle point having searched, make that angle point is capable meets the requirement of column direction symmetry, enter step (12); If search finishes not find two adjacent normal angle points capable, current checkerboard image is processed as abnormal conditions, enter step (13);

Step (12): output image angle point;

Step (13): finish.

Although above-mentioned, by reference to the accompanying drawings the specific embodiment of the present invention is described; but be not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various modifications that creative work can make or distortion still in protection scope of the present invention.

Claims (10)

1. the camera calibration checkerboard image angular-point detection method based on symmetrical analysis, its feature comprises: carry out checkerboard image and read; The image reading is carried out to rim detection; Edge image carries out binaryzation; On the edge image of binaryzation, carry out candidate feature point and detect, obtain candidate feature point; The candidate feature point detecting is carried out to cluster; All candidate feature points in each candidate feature point cluster are carried out to symmetry calculating, obtain unique point set; In unique point set, carry out initial Corner Detection; Take initial angle point as initial search point, estimate other corner location, in the error range of position, the unique point at estimated corner location place is set to angle point; To the angle point searching is capable, judge; Export final angle point.

2. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 1, is characterized in that, concrete steps are as follows:

Step (1): start, read checkerboard image;

Step (2): adopt canny operator to carry out rim detection to image, obtain edge image;

Step (3): edge image carries out binaryzation, obtains the edge image of binaryzation;

Step (4): carry out candidate feature point and detect on the edge image of binaryzation, by the moving window traversal binaryzation edge image that size is w * w is set, if there are 4 intersection points at the edge in the edge of moving window and binaryzation edge image, the central point of moving window are set to candidate feature point, thereby obtain the set of candidate feature point;

Step (5): adopt the method for picture point mark that the candidate feature point detecting is carried out to cluster, obtain the cluster of candidate feature point;

Step (6): adopt ChESS operator to calculate the symmetrical response of all candidate feature points in each candidate feature point cluster;

Step (7): judge according to the symmetrical response of candidate feature point whether this point is unique point, if it is this point is put into unique point set, if not casting out this point;

Step (8): whether the extraction that judges all candidate feature point clusters institute character pair point completes, if completed, enters step (9); If do not completed, return to step (6);

Step (9): in unique point set, carry out initial Corner Detection;

Step (10): take initial angle point as initial search point, estimate other corner location, in the error range of position, the unique point at corner location place is set to angle point; The angle point searching is arranged by row, and claimed that this row angle point is that angle point is capable; The number of contained angle point in checkerboard image a line angle point is set, capable if the angle point number containing in a line angle point searching and identical this row angle point of angle point number of setting are set to normal angle point, if angle point number difference is set to improper angle point capable;

Step (11): whether contain two adjacent normal angle points during the angle point that searched of judgement is capable capable; If it is capable two adjacent normal angle points to be detected, adopt normal angle point capable of the capable alignment of data that carries out of the angle point having searched, make that angle point is capable meets symmetry requirement, enter step (12); If search finishes not find two adjacent normal angle points capable, current checkerboard image is processed as abnormal conditions, enter step (13);

Step (12): output image angle point,

Step (13) finishes.

3. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 2, it is characterized in that, the candidate feature point that the method that adopts picture point to mark in described step (5) obtains detection carries out cluster, and the method that obtains the set of candidate feature point cluster is:

First candidate feature point is marked, compose the identical label of neighboring candidate unique point, will there is the connected domain of the candidate feature point of same numeral as candidate feature point; Minimum connected domain distance is set, and the different connected domains that distance are less than to this setpoint distance are merged into a connected domain, and regard the connected domain after each merging a cluster of candidate feature point as, thereby realize the cluster of candidate feature point.

4. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 2, it is characterized in that, in described step (6), adopt the method for candidate feature point symmetry response in ChESS operator calculated candidate feature points clustering to be:

If candidate feature point set P _fdbe divided into N candidate feature point cluster { s ₁, s ₂..., s _n, cluster s _iin contain m candidate feature point { d ₁, d ₂..., d _m, calculate s _iin the symmetry analog value of each candidate feature point, by s _imiddle symmetrical response is the highest and be greater than 0 candidate feature point and put into unique point set as unique point.

5. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 4, is characterized in that, the formula of the symmetrical response of described calculated candidate unique point is

$R\left(f_i\right)=\underset{j=1}{\overset{m}{\max }}\left(R\left(d_j\right)\right)---\left(1\right)$

Wherein, R (d _j) expression candidate feature point d _jsymmetrical response; R(d _j) computing formula be:

R(d _j)＝SR-DR-MR (2)

Wherein, SR, DR and MR represent respectively overall response value, differential response value and the average response of candidate feature point.Suppose with d _jfor sampling center, it is r that sample radius is set, and sampled point number is n, chooses sampling point set I={I in checkerboard image ₁, I ₂..., I _n, the computing formula of SR, DR and MR is respectively:

$\mathrm{SR}=\underset{k=1}{\overset{n/4}{\mathrm{\Σ}}}|(I_k+I_{k+n/2})-(I_{k+1/4}+I_{k+3n/4})|---\left(3\right)$

$\mathrm{DR}=\underset{k=1}{\overset{n/2}{\mathrm{\Σ}}}|I_k-I_{k+n/2}|---\left(4\right)$

$\mathrm{MR}=|\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{I}_k-\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{I}_k^{\′}|---\left(5\right)$

Wherein, while calculating MR, represent d _jadjacent image point in checkerboard image, gets d _j8 Neighborhood Graph picture point conducts

6. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 2, is characterized in that, the method for carrying out initial Corner Detection in described step (9) is:

Feature points set P _fin with unique point p ₀centered by comprise p ₀in interior any 9 adjacent feature points, distance center unique point p ₀4 nearest unique point p ₁～p ₄for closing on unique point, central feature point p ₀the unique point p of diagonal ₅～p ₈for diagonal angle unique point;

If above-mentioned unique point set P _f9 unique points of middle arbitrary neighborhood meet closes on unique point symmetry, diagonal angle unique point symmetry and closes on the requirement of unique point angular range, these 9 adjacent unique points is defined as to initial angle point.

7. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 6, is characterized in that, described in close on unique point symmetry require be:

Suppose unique point p ₀～p ₈coordinate be respectively (x _i, y _i), 0≤i≤8, if in the error range of position, close on unique point p ₁～p ₄coordinate and p ₀between coordinate, meet following relation:

$\left\{\begin{array}{c}{x}_0=\frac{x_j+x_{j+2}}{2}\\ y_0=\frac{y_j+y_{j+2}}{2}\end{array}\right.---\left(6\right)$

Claim to close on unique point p ₁～p ₄symmetrical about central feature point p0, meet and close on the requirement of unique point symmetry; Wherein, (x ₀, y ₀) centered by the coordinate of unique point p0, (x _j, y _j) represent to close on the coordinate of unique point, j=1,2.

8. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 6, is characterized in that, unique point symmetry requirement in described diagonal angle is:

Suppose that diagonal angle unique point is p _k, 5≤k≤8, the unique point of closing on that it is adjacent is respectively p _iand p _j, 1≤i, j≤4, if in the error range of position p _kcoordinate and p _iand p _jand p ₀between coordinate, meet:

$\left\{\begin{array}{c}{x}_c=\frac{x_i+x_j}{2}\\ y_c=\frac{y_i+y_j}{2}\\ x_k={2x}_c-x_0\\ y_k={2y}_c-y_0\end{array}\right.---\left(9\right)$

Claim diagonal angle unique point p ₅～p ₈meet symmetry requirement; (x _i, y _i), (x _j, y _j), (x ₀, y ₀), (x _k, y _k) represent respectively p _i, p _j, p ₀and p _kcoordinate, (x _c, y _c) for closing on unique point p _iand p _jthe coordinate of central point.

9. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 6, is characterized in that, described in close on unique point angular range require be:

Close on unique point p ₁～p ₄with p ₀between angle can be expressed as:

${\mathrm{\θ}}_{p_i{p}_0{p}_{i+1}}=\arccos \left(\frac{d_1^2+d_2^2-d_3^2}{{2d}_1{d}_2}\right)---\left(10\right)$

Wherein, d ₁, d ₂and d ₃represent respectively p _iwith p ₀, p _i+1with p ₀, p _iwith p _i+1between distance, $d_1=\sqrt{{(x_i-x_0)}^2+{(y_i-y_0)}^2},$ $d_2=\sqrt{{(x_{i+1}-x_0)}^2+{(y_{i+1}-y_0)}^2},$ $d_3=\sqrt{{(x_{i+1}-x_i)}^2+{(y_{i+1}-y_i)}^2},$ (x _i, y _i), (x _i+1, y _i+1), (x ₀, y ₀) represent respectively p _i, p _i+1and p ₀coordinate, when i=4, get p _i+1=p ₁;

If meet:

${\mathrm{\θ}}_{p_i{p}_0{p}_{i+1}}\≥{\mathrm{\θ}}_{\min }---\left(11\right)$

Claim to close on unique point and central feature point meets angle requirement; Wherein, θ _minfor the minimum angle value arranging, get θ _min>=15 °.

10. a kind of camera calibration checkerboard image angular-point detection method based on symmetrical analysis as claimed in claim 2, is characterized in that, the concrete grammar of described step (10) is:

By other corner location of known angle point estimation, comprise following three kinds of situations:

1) if above angle points continuously of 2 and 2 in known a line are estimated other angle point in this row: suppose p ₁and p ₂represent respectively two adjacent corner points of angle point in capable, calculate respectively p ₁about p ₂symmetric points p ₃and p ₂about p ₁symmetric points p ₄position, within the scope of position estimation error, extract P _fmiddle p ₃and p ₄the unique point of position is as this row angle point; Adopt successively the method, realize this angle point capable in the detection of all angle points;

2) capable if the angle point having detected comprises two adjacent normal angle points in capable, according to normal angle point capable estimate other angle point capable in the position of angle point: suppose p ₁₁, p ₁₂and p ₁₃for the angle point in normal angle point capable 1, p ₂₁, p ₂₂and p ₂₃angle point in normal angle point capable 2, calculates respectively p ₁₁about p ₂₁symmetric points p ₃₁, p ₂₁about p ₁₁symmetric points p ₀₁, p ₁₂about p ₂₂symmetric points p ₃₂, p ₂₂about p ₁₂symmetric points p ₀₂, p ₁₃about p ₂₃symmetric points p ₃₃, p ₂₃about p ₁₃symmetric points p ₀₃; P ₀₁, p ₀₂, p ₀₃for the position of angle point in adjacent corner points row 0, p ₃₁, p ₃₂, p ₃₃position for angle point in adjacent corner points row 3; In the error range of position, by unique point set P _fthe unique point of middle corresponding position is set to angle point;

3) if the angle point having detected does not have two adjacent normal angle points capable in capable, according to existing angle point capable estimate other angle points capable in the position of angle point: suppose p ₁, p ₂and p ₃for in the capable H of angle point ₀three adjacent angle points in centre position, H ₁for H ₀adjacent corner points row, p ₄for in H ₁the angle point in centre position, calculates respectively p ₁, p ₂and p ₃about p ₄symmetric points p ₅, p ₆and p ₇, p ₅～p ₇for the new capable H of angle point ₂in corner location, in the error range of position, by P _fat p ₅～p ₇the unique point of position is set to angle point.