CN101794448B - Full automatic calibration method of master-slave camera chain - Google Patents

Abstract

A fully automatic calibration method for a master-slave camera system in the field of image processing technology. The present invention stitches the images automatically obtained during the process of rotating a dynamic camera into a mosaic image, and then automatically extracts and compares the feature points of the image with the master camera image. Matching, so as to obtain the calibration result between the pixel coordinates of the static camera image and the control parameters of the dynamic camera, and realize the automatic calibration of the master-slave camera system. The field of view of the dynamic camera is represented by mosaic stitching images, and the relationship between two different imaging planes is estimated by using SURF feature points. The master-slave camera system can be automatically calibrated only by formulating the motion trajectory of the dynamic camera in advance. , can realize the automatic calibration of the master-slave camera system under the condition of maintaining good calibration accuracy (the error rate is 3%-5%), which saves manpower and time, and reduces the complexity of the calibration process.

Description

Fully automatic calibration method for master-slave camera system

technical field

The invention relates to a method in the technical field of image processing, in particular to a fully automatic calibration method for a master-slave camera system.

Background technique

At present, public places are generally equipped with camera monitoring systems, and a large number of cameras are installed for large coverage areas. With the development of surveillance systems, there is an increasing demand for high-resolution surveillance images. The master-slave camera monitoring system is a solution to this problem. In such a system, one (or more) fixed cameras act as a leader, leading one (or more) dynamic (PTZ) cameras to focus on objects of interest. In this way, a macro surveillance image of a wide area can be obtained, and a high-resolution image of the target can be obtained. The calibration of the master-slave camera system is the solution to this problem. After the calibrated master-slave camera monitoring system, when the master camera captures a target of interest, the dynamic camera can automatically aim at the target to obtain detailed images.

Find through document retrieval to prior art, relevant document is as follows:

1. "A master-slave system to acquire biometric imagery of humans at distance" published by X.Zhou et al. in The First ACM International Workshop on Video Surveillance The master-slave system for obtaining the characteristics of human biological images)" selects some sample points from the main camera image, manually moves the dynamic camera so that the center of the dynamic camera image corresponds to the selected sample points, and then records the coordinates of the sample points and controls parameter. The dynamic camera rotation parameters corresponding to the points on other static camera images can be obtained by interpolating the obtained parameters. However, this technique requires manual adjustment of dynamic camera parameters, which is time-consuming and inconvenient for large-scale application.

2. "Acquiring Multi-Scale Images by Pan-Tilt-Zoom Control and Automatic Multi- Camera Calibration (multi-scale images obtained through PTZ control and multi-camera automatic calibration)", this technology estimates the relationship between the main camera and the dynamic camera by calculating the homography matrix between the two camera images. However, this technology also requires manual adjustment of dynamic camera parameters, which is also time-consuming and inconvenient for large-scale application.

Contents of the invention

Aiming at the above-mentioned deficiencies in the prior art, the present invention proposes a fully automatic calibration method for a master-slave camera system. The present invention stitches the images automatically obtained in the process of rotating the dynamic camera into a mosaic image, and then automatically extracts and matches the feature points of the image with the main camera image, thereby obtaining the relationship between the pixel coordinates of the static camera image and the position parameters of the dynamic camera. Calibration results, to achieve fully automatic calibration of the master-slave camera system.

The present invention is achieved through the following technical solutions, comprising the steps of:

Step 1, pre-set the motion track of the dynamic camera, the dynamic camera automatically rotates along the motion track, and samples a sub-image every time t, and labels and records each sub-image according to the sequence of time collected. The parameters of the dynamic camera rotation when collecting each sub-image, namely: the horizontal rotation angle and the vertical rotation angle, and store each sub-image and its corresponding label together with the corresponding parameters and position information of the dynamic camera rotation.

Step 2: After the dynamic camera rotation is completed, the obtained N sub-images are grouped and arranged according to the shooting positions to obtain K groups of sub-image sets, and the feature points are extracted and summed for the two adjacent sub-images in each group of sub-image sets. Matching processing, to obtain matching feature point pairs, and then to obtain the transformation matrix M between two adjacent sub-images in the same group of sub-image sets.

The grouping arrangement process is to divide the sub-images corresponding to the sub-images whose dynamic camera horizontal rotation angle difference is smaller than the threshold into a group of sub-image sets, thereby obtaining K groups of sub-image sets, and divide the sub-images in each group according to the corresponding The vertical rotation angle of the dynamic camera is arranged.

The feature point extraction and matching process is completed by the SURF method, that is, the fast Hessian extractor extracts the feature points of two adjacent sub-images in each group of sub-image sets, and then obtains the SURF feature vector of the feature points.

The transformation matrix M between two adjacent sub-images in the same group of sub-image sets is obtained by the RANSAC method, specifically:

$\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{ccc}{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}\\ {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 6</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 7</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 8</span>}}\end{array}\right]<span\; class="notranslate">\; *</span>\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\\ \mathrm{<span\; class="notranslate">\; z</span>}\end{array}\right]\mathrm{<span\; class="notranslate">\; or</span>}{\mathrm{<span\; class="notranslate">\; u</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; Mu</span>}<span\; class="notranslate">\; ,</span>$

Among them: (x′, y′, z′) and (x, y, z) are the coordinates of the same feature point in the coordinate system of two adjacent sub-images respectively, and the sub-image coordinate system takes the upper left corner of the image as the origin, The x-axis is positive horizontally to the right, and the y-axis is vertically downward.

Step 3: Perform image mosaic processing on each group of sub-image sets according to the obtained transformation matrix, so that each group of sub-image sets is stitched into a row-stitched image, and a total of K row-stitched images are obtained.

The image splicing process specifically includes: taking the middlemost sub-image in each group of sub-image sets as the reference image of the group of sub-image sets, transforming other sub-images in the group of sub-image sets to the reference image according to the obtained transformation matrix In the image plane coordinate system, weighted grayscale splicing is performed on the pixels in the overlapping area of the image, and the grayscale of the pixels in the non-overlapping area of the image remains unchanged, so that each group of sub-images is stitched into a single image. Stitched images in columns.

The grayscale splicing process with weights is specifically:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; res</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}}<span\; class="notranslate">\; ,</span>$

Wherein: f _res (P) is the pixel value of point P after splicing, f _i (p) is the pixel value of point P in the i-th image, W is the number of images participating in splicing, n is a constant, and d _i is P The shortest distance from the point to the boundary of the i-th image participating in stitching.

Step 4: Taking K series of stitched images as an image set, performing feature point extraction, matching processing, and image stitching processing on them in sequence, so as to obtain a mosaic stitching image.

Step 5: Collect a static image by a static camera, perform feature point extraction and matching processing on the static image and the mosaic stitched image, and further search for feature points using the principle of epipolar geometry to obtain matching features between the static image and the mosaic stitched image point pairs, and make the matched feature points distributed as evenly as possible on the static image and the mosaic image.

The epipolar line geometry principle is: the corresponding point of a point in one image on another image is located on the corresponding epipolar line.

The further search for feature points is to make the corresponding points of the feature points of the static image on the mosaic stitched image located on the corresponding epipolar line, thereby increasing the number of feature points.

In step six, the global calibration process is performed on the static image and the mosaic image, so as to obtain the mapping relationship between the static image and the mosaic image, that is, the calibration relationship between the master and slave cameras.

The global calibration process includes the steps of:

1) In the vicinity of any point P _s (x _s , y _s ) in the static image, search for static image feature points, _NR (P _s ) is all static image feature points whose distance from point P _s is less than or equal to R A set consisting of:

$N_R\left(P_{\mathrm{the\; s}}\right)=\{(P_{\mathrm{the\; s}}^1,r_1),(P_{\mathrm{the\; s}}^2,r_2),...\},$ Among them: P _s ⁱ is the feature point whose distance P _s is r _i ;

2) Find the mosaic image feature point P _d ⁱ corresponding to each static image feature point P _s ⁱ in the _NR (P _s ) set, and obtain the sub-image including P _d ⁱ , and then from the sub-image including P _d ⁱ Select a sub-image I _r ⁱ that P _d ⁱ is closest to the center of the sub-image where it is located;

3) Find the label of the sub-image I _r ⁱ in all sub-image sets, and obtain the position parameter S _i of the dynamic camera when shooting the sub-image I _r ⁱ ;

4) Perform interpolation processing on the position parameter S _i of the dynamic camera to obtain the corresponding relationship between any point P _s (x _s , y _s ) in the static image and the position parameter of the dynamic camera, specifically:

S＝S ₁ *f ₁ (r ₁ , f ₂ ,...r _n )+S ₂ *f ₂ (r ₁ , r ₂ ,...r _n )+...+S _n *f _n ( r ₁ , r ₂ ,...r _n ),

Among them: f _i is the interpolation function.

Compared with the prior art, the beneficial effect of the present invention is: the field of view range of the dynamic camera is represented by the mosaic mosaic image, and the relationship between two different imaging planes is estimated by using SURF feature points, only the dynamic camera needs to be prepared in advance. The master-slave camera system can be calibrated automatically, and the master-slave camera system can be automatically calibrated under the condition of maintaining good calibration accuracy (the error rate is 3%-5%), which saves manpower and time , reducing the complexity of the calibration process.

Description of drawings

Fig. 1 is the schematic diagram of embodiment dynamic camera rotation track;

Among them: (a) is the position of the dynamic camera when shooting 64 sub-images; (b) is a number of sub-images taken by the dynamic camera.

Fig. 2 is a schematic diagram of weighted gray-level mosaic processing.

Fig. 3 is the mosaic mosaic image obtained in the embodiment.

Fig. 4 is the matching characteristic point that embodiment obtains;

Among them: (a) is the matching feature point obtained by simply using the SURF method; (b) is the matching feature point obtained by further using the epipolar geometric principle.

FIG. 5 is a schematic diagram of global calibration processing.

Fig. 6 is the calibration result of embodiment;

Among them: (a) is an image captured by a static camera; (b) is an image captured by a dynamic camera relative to (a); (c) is another image captured by a static camera; (d) is an image relative to (c ) images captured by a static camera.

Fig. 7 is the schematic diagram of embodiment calibration error rate;

Wherein: (a) is a schematic diagram of the error rate in the X direction of the embodiment; (b) is a schematic diagram of the error rate in the Y direction of the embodiment.

Detailed ways

The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings: the present embodiment is implemented on the premise of the technical solution of the present invention, and detailed implementation methods and processes are provided, but the protection scope of the present invention is not limited to the following implementations example.

Example

In this embodiment, a static camera and a dynamic camera installed in different places in a certain room are calibrated. The resolution of static camera and dynamic camera is 320*240.

This embodiment includes the following steps:

Step 1, pre-set the motion track of the dynamic camera as a zigzag, the dynamic camera automatically rotates along the given motion track, and samples a sub-image every time t, and at the same time, according to the time sequence of acquisition, for each The sub-image label records the parameters of the dynamic camera rotation each time the image is captured, namely: the horizontal rotation angle and the vertical rotation angle, and combines each sub-image and its corresponding label with the corresponding dynamic camera rotation parameters and position information. and store it.

In the present embodiment, the motion track of the dynamic camera is shown in the arrow in Fig. 1 (a), and a sub-image is sampled at the dots in Fig. 2 respectively, and 64 sub-images are obtained altogether, and some sub-images obtained are as shown in Fig. 1 ( b) as shown.

Step 2: After the dynamic camera rotation is completed, the obtained N sub-images are grouped and arranged according to the shooting position to obtain K groups of sub-image sets, and feature point extraction and matching are performed on two adjacent sub-images in each group of sub-image sets Processing to obtain matching feature point pairs, and then obtain the transformation matrix M between two adjacent sub-images in the same group of sub-image sets.

The grouping arrangement process is to divide the 64 sub-images into 8 groups according to the horizontal rotation angle of the camera, each group contains 8 sub-images, and the sub-images in each group are divided into 8 groups according to the vertical rotation angle of the corresponding dynamic camera. Sort from smallest to largest.

The feature point extraction and matching process is completed by the SURF method, that is, the fast Hessian extractor extracts the feature points of two adjacent sub-images in each group of sub-image sets, and then obtains the SURF feature vector of the feature points.

The transformation matrix M between two adjacent sub-images in the same group of sub-image sets is obtained by the RANSAC method, specifically:

$\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; z</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{ccc}{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}\\ {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 6</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 7</span>}}& {\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 8</span>}}\end{array}\right]<span\; class="notranslate">\; *</span>\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\\ \mathrm{<span\; class="notranslate">\; z</span>}\end{array}\right]\mathrm{<span\; class="notranslate">\; or</span>}{\mathrm{<span\; class="notranslate">\; u</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; Mu</span>}<span\; class="notranslate">\; ,</span>$

Among them: (x′, y′, z′) and (x, y, z) are the coordinates of the same feature point in the coordinate system of two adjacent sub-images respectively, and the sub-image coordinate system is based on the upper left corner of the image as the origin , The x-axis is positive when it is horizontal to the right, and the positive is established when the y-axis is vertically downward.

Step 3: According to the obtained transformation matrix, image stitching is performed on each group of sub-image sets, so that each group of sub-image sets is stitched into a column-stitched image, and a total of 8 column-stitched images are obtained.

The image splicing process specifically includes: taking the middlemost sub-image (the fifth sub-image in this embodiment) in each group of sub-image sets as the reference image of the group of sub-image sets, The other sub-images are transformed into the reference image plane coordinate system according to the obtained transformation matrix, and the pixels in the overlapping areas of the images are processed with weighted grayscale splicing, and the grayscales of the pixels in the non-overlapping areas of the images remain unchanged , so that each group of sub-image sets is stitched into a column-stitched image.

As shown in Figure 2, the gray-scale mosaic processing with weights is specifically:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; res</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 8</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 8</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}}<span\; class="notranslate">\; ,</span>$

Among them: f _res (P) is the pixel value of point P after splicing, f _i (p) is the pixel value of point P in the i-th image, n is a constant, and d _i is from point P to the i-th frame involved in splicing The shortest distance to the image border.

n is 3 in this embodiment.

Step 4: Take the 8 column-stitched images as an image set, and perform feature point extraction, matching processing, and image stitching processing on them in sequence according to the methods in Step 2 and Step 3, so as to obtain a mosaic stitching image.

The mosaic stitched image obtained in this embodiment is shown in FIG. 3 , where the white line is the boundary of each sub-image.

Step 5: Collect a static image by a static camera, perform feature point extraction and matching processing on the static image and the mosaic stitched image, and further search for feature points using the principle of epipolar geometry to obtain matching features between the static image and the mosaic stitched image Point pairs, and make the matching feature points distributed on the two images as evenly as possible.

The epipolar line geometry principle is: the corresponding point of a point in one image on another image is located on the corresponding epipolar line.

In this embodiment, the matching feature points obtained simply by using the SURF method are shown in FIG. 4( a ), and the matching feature points obtained by further using the epipolar geometric principle are shown in FIG. 4 ( b ).

In step six, the global calibration process is performed on the static image and the mosaic image, so as to obtain the mapping relationship between the static image and the mosaic image, that is, the calibration relationship between the master and slave cameras.

As shown in Figure 5, the described global calibration process includes the following steps:

1) In the vicinity of any point P _s (x _s , y _s ) in the static image, search for the static image SURF feature points, _NR (P _s ) is all the static image SURF whose distance from point P _s is less than or equal to R A collection of feature points:

$N_R\left(P_{\mathrm{the\; s}}\right)=\{(P_{\mathrm{the\; s}}^1,r_1),(P_{\mathrm{the\; s}}^2,r_2),...\},$ Among them: P _s ⁱ is the feature point whose distance P _s is r _i ;

2) Find the mosaic image feature point P _d ⁱ corresponding to each static image SURF feature point P _s ⁱ in the _NR (P _s ) set, obtain the sub-image including P _d ⁱ , and then obtain the sub-image including P _d ⁱ Select the sub-image I _r ⁱ that P _d ⁱ is closest to the center of the sub-image;

3) Find the label of the sub-image I _r ⁱ in all sub-image sets, and obtain the position S _i (α _i , β _i , Z _i )i=1, 2, where the dynamic camera is when shooting the sub-image I _r ⁱ 3...;

4) Perform interpolation processing on S _i (α _i , β _i , Z _i ) to obtain the corresponding relationship between any point P _s (x _s , y _s ) in the static image and the position parameters of the dynamic camera, specifically:

S=S ₁ *f ₁ (r ₁ , r ₂ ,...r _n )+S ₂ *f ₂ (r ₁ ,r ₂ ,...r _n )+...+S _n *f _n ( r ₁ , r ₂ ,...r _n ),

Among them: f _i is the interpolation function.

In this embodiment, the interpolation function is specifically: $f_i(r_1,r_2,...r_{\mathrm{no}})=r_i/\sqrt{\underset{i=1}{\overset{\mathrm{no}}{\mathrm{\&Sigma;}}}{r}_i^2}.$

Use a static camera to shoot an image as shown in Figure 6(a), and use the method of this embodiment to calibrate the dynamic camera for the object of interest (person) in Figure 6(a), so that the image obtained by rotating the dynamic camera is shown in Figure 6(a). 6(b); similarly, the static camera captures another image as shown in Figure 6(c), and the corresponding dynamic camera image is shown in Figure 6(d).

This embodiment selects any 8 points in the static image, and calculates the corresponding rotation angle of the dynamic camera according to the obtained calibration relationship between the master and slave cameras. After rotating the dynamic camera, calculate the point corresponding to the center of the dynamic camera image and the selected point at The distance in the X and Y directions, so that according to the ratio of the distance in the X and Y directions to the resolution of the dynamic camera, the error rate of the master-slave camera calibration is obtained. The error rate in the X direction obtained in this embodiment is shown in Figure 7 (a) As shown, the error rate in the Y direction is shown in FIG. 7(b), and it can be seen from FIG. 7 that the method of this embodiment is simple and the calibration accuracy is high.

Claims (7)

1. the full automatic calibration method of a master-slave camera chain is characterized in that, may further comprise the steps:

Step 1; Preestablish the movement locus of dynamic camera; Dynamic camera rotates along this movement locus automatically; And per interval t width of cloth subimage of sampling; Simultaneously by the time sequencing that collects; To each width of cloth subimage label, the parameter that dynamic camera rotates during the every width of cloth subimage of record acquisition, and every width of cloth subimage and corresponding label thereof stored with parameter and positional information that corresponding dynamic camera rotates;

Step 2; After the dynamic camera rotation finishes; The N width of cloth subimage that obtains is carried out arranged in groups by the camera site to be handled; Obtain K group set of sub-images; Two width of cloth subimages to adjacent in every group of set of sub-images carry out feature point extraction and matching treatment; It is right to obtain matched feature points, and then obtains the transform matrix M between two adjacent in same group of set of sub-images width of cloth subimages;

Step 3 based on the transformation matrix that obtains, is carried out image mosaic to every group of set of sub-images and is handled, thereby every group of set of sub-images is spliced into a width of cloth row stitching image, obtains K width of cloth row stitching image altogether;

Step 4 as an image collection, is carried out K width of cloth row stitching image successively feature point extraction, matching treatment and image mosaic to it and is handled, thereby obtain a width of cloth mosaic stitching image;

Step 5; Gather a width of cloth still image by still camera; This still image and mosaic stitching image are carried out feature point extraction and matching treatment; And utilize the further search characteristics point of polar curve geometrical principle; The matched feature points that obtains between still image and mosaic stitching image is right, and the characteristic point of coupling is evenly distributed on still image and the mosaic stitching image as far as possible;

The described further search characteristics point of polar curve geometrical principle that utilizes is that the characteristic point of still image is positioned on the corresponding polar curve in the corresponding points on the mosaic stitching image, thereby increases the quantity of characteristic point.

Step 6 is carried out the global calibration processing to still image and mosaic stitching image, thereby is obtained the mapping relations between still image and mosaic stitching image, be i.e. demarcation between principal and subordinate's video camera relation;

Described global calibration is handled, and comprises that step is:

1) any some P in still image _s(x _s, y _s) in the adjacent domain, search still image SURF characteristic point, N _R(P _s) be range points P _sDistance is smaller or equal to the set of all still image SURF characteristic points formations of R, that is:

Wherein

It is distance P _sBe r _iCharacteristic point;

2) find N _R(P _s) each still image characteristic point in the set

Corresponding mosaic image characteristic point

Comprised

Subimage, and then from comprising

Subimage in select

Belong to the subimage at subimage center near it

3) search subimage

Label in all set of sub-images obtains dynamic camera and is taking subimage The time residing location parameter S _i

4) to the location parameter S of dynamic camera _iCarry out interpolation processing, obtain any point P in the still image _s(x _s, y _s) with the corresponding relation of the location parameter of dynamic camera be:

S＝S ₁*f ₁(r ₁，r ₂，...r _n)+S ₂*f ₂(r ₁，r ₂，...r _n)+...+S _n*f _n(r ₁，r ₂，...r _n)，

Wherein: f _iBe interpolating function.

2. the full automatic calibration method of master-slave camera chain according to claim 1 is characterized in that, the parameter that the dynamic camera described in the step 1 rotates is meant: horizontally rotate angle and vertical rotation angle.

3. the full automatic calibration method of master-slave camera chain according to claim 1; It is characterized in that; It is that the dynamic camera that subimage is corresponding horizontally rotates differential seat angle and is divided into one group of set of sub-images less than the subimage of threshold value that arranged in groups described in the step 2 is handled; Thereby obtain K group set of sub-images, and the subimage in every group is arranged according to the vertical rotation angular dimension of the dynamic camera of correspondence.

4. the full automatic calibration method of master-slave camera chain according to claim 1; It is characterized in that; Feature point extraction described in the step 2 and matching treatment are to be extracted the characteristic point of adjacent two width of cloth subimages in every group of set of sub-images by quick Hessian extractor, obtain the SURF characteristic vector of characteristic point again.

5. the full automatic calibration method of master-slave camera chain according to claim 1 is characterized in that, the transform matrix M in the same group of set of sub-images described in the step 2 between adjacent two width of cloth subimages is:

Wherein: (x ', y ', z ') ⊥ is with (z) ⊥ is respectively the coordinate of same characteristic point in two adjacent width of cloth subimage coordinate systems for x, y, and the subimage coordinate system is an initial point with the image upper left corner, and the x axle horizontal is to the right for just, and the y axle is what just setting up vertically downward.

6. the full automatic calibration method of master-slave camera chain according to claim 1; It is characterized in that; Image mosaic described in the step 3 is handled: with the benchmark image of subimage middle in every group of set of sub-images for this group set of sub-images; Other subimages in this group set of sub-images are converted in the benchmark image plane coordinate system according to the transformation matrix that obtains; The pixel in the coincidence of image zone is carried out the heavy gray scale splicing of cum rights handle, the gray scale of the pixel in the non-coincidence zone of image is remained unchanged, thereby every group of set of sub-images is spliced into a width of cloth row stitching image.

7. the full automatic calibration method of master-slave camera chain according to claim 6 is characterized in that, the gray scale splicing that described cum rights is heavy is handled and is:

Wherein: f _Res(P) be the pixel value that splicing back P is ordered, f _i(p) be the pixel value that P is ordered in the i width of cloth image, W is for participating in spliced image quantity, and n is a constant, d _iBe the beeline of P point to the i width of cloth image boundary of participating in splicing.

Images