CN105631844A - Image camera calibration method - Google Patents

Abstract

The invention relates to an image camera calibration method. The image camera calibration method comprises the steps of obtaining a nonlinear imaging model in which distortion is introduced through determining the geometric imaging relationship of an image camera and introducing image camera distortion correction, and furthermore determining all to-be-calculated image camera parameters; acquiring a plurality of calibration board images in different visual angles as references, extracting image coordinate information in the calibration board images, introducing the to-be-calculated image camera parameters as objects into a differential evolution particle swarm optimization for performing optimization solution, thereby calibrating all image camera parameters, and furthermore realizing mapping from a three-dimensional spatial position to a two-dimensional image, thereby finishing calibration of a coordinate which corresponds with the two-dimensional image. The image camera calibration method has advantages of high precision and high stability.

Description

A kind of camera marking method

Technical field

The present invention relates to technical field of computer vision, particularly to a kind of camera marking method.

Background technology

Along with the development of science and technology updates, computer vision technique is widely used in daily life and commercial production. The research purpose of computer vision is exactly the ability making computer have by two dimensional image cognition three-dimensional environment information. They, by not only causing a machine to the geological information of object in perception three-dimensional environment, including its shape, position, attitude, motion etc., and can be described, store, identify and understand by this ability.

The image information that computer vision system obtains from video camera, calculates in three-dimensional environment the geological informations such as the position of object, shape, and thus 3 d objects reconstruction. On image, the brightness of every bit reflects the intensity of space object surface point reflection light, and this position on image is relevant with the geometric position of space object surface respective point. The mutual relation of these positions, is determined by video camera imaging geometric model. The parameter of this geometric model is called camera parameters, and these parameters must be determined with calculating by test, and test is called camera calibration with the process calculated. Technology for computer vision realizes, and camera calibration is most important, and it is the basis of computer vision technique, is key and the steps necessary of two dimensional image acquisition three-dimensional spatial information. No matter being in image measurement or machine vision applications, the demarcation of camera parameters is all that very the key link, the precision of its calibration result and the stability of algorithm directly affect the accuracy of camera operation generation result.

At present, camera calibration theory is very ripe, also in succession proposes a lot of scaling method. Can be divided mainly into traditional camera marking method and camera self-calibration method. Tradition scaling method stated accuracy is high, but needs specific scaling reference; Self-calibrating method does not rely on scaling reference, but calibration result is relatively unstable. As far back as early 1970s, Abdel-Aziz and Karara proposes the camera marking method of direct linear transformation paper " in close-range photogrammetry direct linear transformation (Directlineartransformationfromcomparatorcoordinatesintoo bjectspacecoordinatesinclose-rangephotogrammetry) " from comparator coordinate to object space coordinate, establish video camera imaging geometric linear model, can directly obtain imaging model parameter by linear equation. But this linear model does not consider the distortion of video camera, and in actual applications, video camera often exists distortion, so can produce bigger error. In camera calibration subsequently develops, people start the distortion taken into full account existing for video camera. Wherein representative is the Zhang Zhengyou method that a kind of orthogonality condition utilizing spin matrix of proposition and nonlinear optimization carry out camera calibration in paper " new technique (Aflexiblenewtechniqueforcameracalibration) of a kind of camera calibration flexibly ", the method introduces camera lens distortion, plane template is used to make reference thing, use flexibly, very simply. But the method only considered first two of radial distortion for the distortion of video camera, consider its consequent and video camera tangential distortion etc., however it remains certain error. Especially in actual applications, when requiring higher to stated accuracy, there is significant limitation.

Along with the research that deepens continuously of scaling method, and the proposition in succession of some intelligent algorithms and extensive use, intellectual learning algorithm is applied in camera calibration and also achieves significant effect and superiority. K.Deep et al. proposes the camera marking method based on particle cluster algorithm in paper " stereo camera based on particle group optimizing demarcates (Stereocameracalibrationusingparticleswarmoptimization) ", utilize particle cluster algorithm ability of searching optimum, camera parameters can be well carried out and solve. But, there is certain error in the limitation due to intelligent algorithm itself, it is easy to be trapped in the problem such as Premature Convergence or inefficiency, it is impossible to find optimal solution, does not reach the precision solved. Therefore, the connected applications of many algorithms is selected, it is possible to effectively avoid the defect of single algorithm, better find optimal solution. Along with the deep development of intelligent algorithm, the connected applications of algorithm too increases the complexity of demarcation, makes troubles in actual applications. Therefore find the effective algorithm of simple and stable and be applied to camera calibration, there is critically important Research Significance and practical value.

Summary of the invention

In order to realize the three dimensions mapping to two-dimensional space, state three-dimensional geological information as two-dimensional image information, give and represent intuitively, complete visual representation task, and ensure the precision of mark, considering that engineer applied meets simple accurately completing fast and effectively as far as possible and demarcates, the present invention proposes a kind of camera marking method based on differential evolution particle cluster algorithm (DEPSO) simultaneously.

In order to achieve the above object, the technical scheme is that a kind of camera marking method of offer, it comprises procedure below: determines video camera geometry imaging relations, and introduces distortion of camera correction, obtain introducing the nonlinear imaging model of distortion, and then determine all camera parameters to be solved; Gather the scaling board image of several different visual angles as reference substance, extract image coordinate information therein, and with camera parameters to be solved for object, substitution differential evolution particle cluster algorithm carries out optimizing solve, to calibrate all of camera parameters, and then realize the three-dimensional space position mapping to two dimensional image, complete the mark of two dimensional image respective coordinates.

The present invention passes through a kind of camera marking method based on differential evolution particle cluster algorithm, have an advantage in that: owing to the realization of differential evolution algorithm and particle cluster algorithm is simple, adjustable parameter is few, than general traditional intelligence algorithms such as genetic algorithms, there is higher global optimization ability, and the precision of optimizing is relatively higher, has better stability. The combination of two kinds of algorithms, utilizes the intersection of differential evolution algorithm, and make a variation selection strategy, it is to avoid particle cluster algorithm Premature Convergence is easily absorbed in the defect of local optimum, thus for the research of camera calibration technology.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of camera marking method of the present invention;

Fig. 2 is the coordinate schematic diagram of nonlinear imaging model in camera marking method of the present invention;

Fig. 3 is the schematic diagram of scaling board image in camera marking method of the present invention.

Detailed description of the invention

The present invention provides a kind of camera marking method based on differential evolution particle cluster algorithm, combines differential evolution algorithm and particle cluster algorithm, and its concrete implementation flow process is as shown in Figure 1. Wherein, video camera geometry imaging relations is the basis of camera calibration, it is determined that imaging model, just can determine that camera interior and exterior parameter. Further, the present invention takes into full account radial distortion and the tangential distortion of video camera, introduce distortion of camera correction, it is determined that distortion parameter, so that it is determined that all of video camera solve parameter. Adopting several scaling board images is reference substance, extracts image coordinate information, then with required parameter for object, carries out optimizing and solve, thus calibrating all camera parameters to be solved in substitution differential evolution particle cluster algorithm. The three-dimensional space position mapping to two dimensional image can be realized according to required parameter, it is determined that the respective coordinates of two dimensional image, complete mark task, so that it is determined that the effectiveness of scaling method of the present invention and accuracy.

Determine that video camera solves parameter

In order to describe the geometry imaging relations of space three-dimensional information and two-dimensional image information, it is necessary first to camera is carried out mathematical modeling, it is determined that the imaging relations of video camera, just so can determine that the parameter of video camera and corresponding method for solving. Camera imaging model includes linear model and nonlinear model. Linear model is general image pinhole model, according to pinhole imaging system principle, sets up picture point and the geometry site of homologue surface spatial point.

In actual applications, due to stereo visual system many employings non-metric camera, its physique results in various distortion so that actual image point position can offset ideal image point position, causes the collinear relationship between picture point, projection centre and corresponding spatial point to be damaged; And, error due to incident ray error of refraction when by each lens and CCD plane lattice position, in image acquisition process, the situation of the local appearance distortion that range image center is more remote is more serious, in this case linear model can not describe imaging geometry exactly, so needing to adopt nonlinear distortion, namely setting up distortion correction model (nonlinear model) and completing.

Based on, on the basis of linear model, adding nonlinear distortion varying model in the inventive method, take into full account the distortion effects of video camera, introduce the maximum radial distortion of distortion effects and tangential distortion. The imaging geometry of video camera is determined with this imaging model, so that it is determined that the parameter all to be solved of video camera.

The nonlinear imaging model that the present invention adopts, relates to the coordinate system shown in Fig. 2:

(1) the image coordinate system o represented in units of pixel_iUv: initial point o_iBeing positioned at the upper left corner of ccd image plane, u axle and v axle and be respectively parallel to image plane pixel row and pixel column, (u v) represents object point P coordinate under the image coordinate system in units of pixel.

(2) the image coordinate system oxy represented with physical unit (such as mm): initial point (i.e. principal point o) is positioned at the intersection point of camera optical axis and ccd image plane, x-axis and y-axis are respectively parallel to u axle and v axle. p_u(x_u,y_u) represent the image coordinate of object point P, p under desirable pin-hole model_d(x_d,y_d) represent the real image coordinate of object point P under the pin-hole model considering lens distortion.

(3) camera coordinate system o_cx_cy_cz_c: initial point o_cIt is positioned at the photocentre of camera lens, x_cAxle, y_cAxle is respectively parallel to x-axis and y-axis, z_cAxle is video camera primary optical axis, and it is vertical with the plane of delineation. o_cO is the effective focal length f of video camera, (x_c,y_c,z_c) represent object point P coordinate under camera coordinate system.

(4) world coordinate system o_wx_wy_wz_w: the frame of reference suitably selected in the environment, (x_w,y_w,z_w) represent object point P coordinate under world coordinate system.

For radial distortion, the distortion at imaging plane center is 0, along with moving to edge, distorts increasingly severe. Quantitative description can be carried out by the first few items of the Taylor series expansion around r=0 position. For general video camera, we generally use front two term coefficient, and wherein Section 1 is generally k₁, Section 2 is k₂. The video camera that distortion is very big, such as fish-eye lens, it is possible to use the 3rd radial distortion item k₃. Adopt these three distortion factors to carry out distortion herein and adjust as follows:

$\begin{array}{c}{x}_u=x_d(1+k_1{r}^2+k_2{r}^4+k_3{r}^6)\\ y_u=y_d(1+k_1{r}^2+k_2{r}^4+k_3{r}^6)\end{array}$

For tangential distortion, this distortion is to produce owing to the defect in lens manufacture makes lens itself and the plane of delineation not parallel. Two parameter p are commonly used in tangential distortion₁And p₂Describe as follows:

$\left\{\begin{array}{c}{x}_u=x_d+[2p_1{y}_d+p_2(r^2+2x_d^2)]\\ y_u=y_d+[p_1(r^2+2y_d^2)+2p_2{x}_d]\end{array}\right.$

Wherein,Described by the above-mentioned definition to two kinds of distortion, in calibration process, consider the impact of two kinds of distortion simultaneously, through arranging, it is possible to the calibration result not distorted is as follows:

$\left[\begin{array}{c}{x}_u\\ y_u\end{array}\right]=[1+k_1(x_d^2+y_d^2)+k_2{(x_d^2+y_d^2)}^2+k_3{(x_d^2+y_d^2)}^3]\left[\begin{array}{c}{x}_d\\ y_d\end{array}\right]+\left[\begin{array}{c}2p_1{x}_d{y}_d+p_2(3x_d^2+y_d^2)\\ p_1(x_d^2+3y_d^2)+2p_2{x}_d{y}_d\end{array}\right]$

By the above-mentioned analysis to camera model, adopt the nonlinear model introducing distortion herein. In actual applications, carry out distortion correction, effectively reduce because of the brought impact that distorts, it is ensured that the accuracy of calibration result, there is critically important researching value.

Algorithm design application calibrating parameters

By the creation analysis to above-mentioned camera model, with camera parameters for optimization aim, differential evolution particle cluster algorithm is applied to during camera calibration solves. The inventive method is to gather the scaling board image of several different visual angles as reference substance, it is provided that image coordinate information, substitutes into during algorithm iteration solves, finds optimum camera parameters. Uncalibrated image used by the present invention is the gridiron pattern that classic black is alternate, as shown in Figure 3. Being provided with 8 �� 10 totally 80 angle points in the present embodiment, when adopting unequal ranks dimension, this unequal pattern to make chessboard Corner Detection, direction can uniquely be determined, it is ensured that the angle point found correctly arranges.

In Fig. 3, chequered with black and white point is angle point. Gridiron pattern is fixed on one flat plate, as calibrating template, at focal length of camera within sweep of the eye, suitably changes the horizontally or vertically orientation angle of calibrating template, gather the scaling board image of some width different visual angles. By Corner Detection Algorithm (not being the emphasis of the inventive method, be not described further), detect the two-dimensional image coordinate that each angle point is corresponding, provide two-dimensional coordinate information for parameter calibration.

If what collect has n width plane template image, having m fixed point (i.e. angle point) in each image, each fixed point is equal in magnitude, is in same outside noise environment. Set up object function as follows:

$f_{\mathrm{obj}}=\min \underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}\left|\right|{\hat{p}}_{\mathrm{ij}}-p(M_A,k_1,k_2,k_3,p_1,p_2,R_i,T_i,P_j)\left|\right|$

Wherein,It it is the pixel coordinate of jth fixed point in the i-th width image. M_AFor intrinsic parameters of the camera battle array, mainly comprise (f_x,f_y,c_x,c_y) these four parameters, wherein f_x=f/d_x, f_y=f/d_yBeing the scale factor of camera image plane horizontal direction x-axis and vertical direction y-axis respectively, f is the focal length of video camera, d_xAnd d_yBe respectively in image plane adjacent two picture points in the horizontal direction with the physical distance in vertical direction, c_xAnd c_yCamera lens center respectively with optical axis in the horizontal direction with the side-play amount of vertical direction.

R_i, T_iIt is the outer parameter that the i-th width image is corresponding, i.e. spin matrix and translation matrix respectively. Outer parameter represents video camera position in world coordinate system and direction, is the relation between unified camera coordinate system and world coordinate system, and wherein spin matrix is made up of three vectors, corresponding to each axial deflection angle tolerance of three dimensions; When translation matrix is used for representing the initial point that the initial point of camera coordinate system moves to world coordinate system, the side-play amount between the two coordinate system.

As described above, k₁, k₂And k₃It is coefficient of radial distortion, p₁And p₂It it is tangential distortion coefficient. P (M_A,k₁,k₂,k₃,p₁,p₂,R_i,T_i,P_j) it is the world coordinates (world coordinates is three dimensional space coordinate) of jth point in the i-th width image, P_jIt is by demarcating the pixel coordinate (pixel coordinate is the image coordinate in units of pixel) tried to achieve.

This object function is a typical non-linear least square problem, utilizes algorithm to be herein optimized and solves, can obtain the optimum stable solution of camera parameters. Specific algorithm application calibration process is as follows:

Step 1, initialization population: in allowed band, randomly generate position and the speed of N number of particle, and set the bound of particle rapidity. In order to avoid algorithm initial stage search capability is strong, the shortcoming that later stage search capability is weak, the parameter that arranges in algorithm is used uniformly across dynamic adjustable strategies, strengthens the search capability of particle.

Allowed band generally takes experience range, different optimization problems, and difference is treated, and scope too conference reduces convergence of algorithm speed. The present invention advises allowed band is set smaller than 100. But, it can be adjusted by other different embodiments according to practical situation. Particle randomly generate be initialize time self-defining, how many numbers is determined according to optimization problem, and these particles are the individualities in algorithm, the solution that namely algorithm iteration computing generates every time.

The dynamic adjustable strategies that parameter is adopted, refer in interative computation, the parameter in algorithm is made all to do corresponding adjustment after iteration each time, namely parameter value can change along with iteration each time, not being uniquely constant value, generally understand maximum and the minima of custom algorithm parameter, namely the change of this value is within the scope of this, general initial stage parameter value is relatively big, and later stage parameter value is less. Iteration can arrange fixing iterations, starts interative computation early stage and is the initial stage, and along with computing iteration, until terminating, iteration below is the later stage.

Step 2, choose fitness function: with the distance of required pixel and actual pixels point for target, as fitness evaluation standard, calculate adaptive value and obtain initializing individual extreme value and the global extremum of colony. Fitness function definition is as follows:

$f\left(\right)=\min \underset{i=1}{\overset{m}{\mathrm{\Σ}}}\sqrt{{(u_i-x_i)}^2+{(v_i-y_i)}^2}$

The position of step 3, more new particle and speed: be updated according to the algorithm iteration formula speed to each particle and position; Particle cluster algorithm iterative formula is:

$v_{\mathrm{id}}^{k+1}={\mathrm{wv}}_{\mathrm{id}}^k+c_1\mathrm{rand}\left(\right)(\mathrm{pbest}-x_{\mathrm{id}}^k)+c_2\mathrm{rand}\left(\right)(\mathrm{gbest}-x_{\mathrm{id}}^k)$

$x_{\mathrm{id}}^{k+1}=x_{\mathrm{id}}^k+v_{\mathrm{id}}^{k+1}$

Wherein: k is iterations; If the information D dimensional vector of particle i represents, positional representation is X_i=(x_i1,x_i2,...,x_iD)^T, speed is V_i=(v_i1,v_i2,...,v_iD)^T. W is inertia weight, affects the global and local search capability of algorithm; Rand () is the random number between [0,1]; Pbest is the particle i position (i.e. coordinate) at the d individual extreme point tieed up; Gbest is the whole population position at the d global extremum point tieed up.

c₁, c₂It it is accelerated factor (or claiming Studying factors), (" best " is to pass judgment on fitness to be respectively regulated to the overall situation preferably particle and the best particle of individuality, namely refer to that fitness is the highest) direction flight maximum step-length, if it is too little, then particle is possibly remote from target area, if too big, then can cause flying over target area. Suitable c₁, c₂Rapid convergence can be added and be not easily absorbed in local optimum, generally making c₁=c₂=2.

Step 4, selection update: adopt the selection strategy of differential evolution, the adaptive value of particle after renewal and the adaptive value updating front particle are compared, the particle selecting fitness high (fitness function is required pixel and the distance of actual pixels point, then distance is more little, and namely fitness is more high) updates position;

Selection operation operator is as follows:

Wherein, definition intersection vector U_i=(u_i1,u_i2,��,u_iD)^T, this intersection vector is to carry out computing in intersection operation below, and the selection in this step is to be selected by fitness.

Step 5, the operation that intersects: increase the multiformity of colony, it is ensured that the defect individual that fitness is high; Definition intersection vector U_i=(u_i1,u_i2,��,u_iD)^T, the crossover operator of intersection operation is:

In formula, j=1,2 ..., N, j_rand�� [1, N]. CR is cross-over control parameter, and general value is between [0,1], and its value is more big, occurs the probability intersected more big, and the multiformity of population is also more big.

Step 6, mutation operation: (more high more good for standard with fitness by comparing fitness, what compare is all of individuality), low fitness individuality is made to make a variation with greater probability, so be conducive to producing excellent pattern, ensure that the existence that advantage is individual, define excellent colony of new generation; The mutation operator of mutation operation computing is:

v_i,G=x_r1,G+F(x_r2,G-x_r3,G)

In formula, r1, r2, r3 is the integer being arbitrarily not mutually equal between [1, N]. F is variance control parameter, namely controls the amplification of difference, and general value is between [0,2]. Low fitness, the distance being required pixel and actual pixels point is relatively big, and the individuality of this solution is low fitness. Low fitness is made a variation individual by mutation operator, again through cross selection, produces new individuality, form new excellent colony particle.

Step 7, update the extreme value of (updating according to above-mentioned iterative formula) colony: the adaptive value according to colony of a new generation, update the individual extreme value of colony and global extremum;

Step 8, judge whether to meet end condition: if reaching maximum iteration time, then end loop output result, otherwise forwards step 3 to and continues executing with.

Two dimensional image coordinates logo

Solved by the application of algorithm, calibrate inside and outside parameter and the distortion factor of video camera. The calibration result of one example is as follows:

And based on this parameter, it is possible to carry out the coordinates logo of two dimensional image. According to model transformation relation, by three dimensional space coordinate (x_w,y_w,z_w), solve corresponding two dimensional image coordinate (u, v), and with processed the image coordinate obtained by real imageCompare, verify that this invention scaling method tries to achieve reliability and the accuracy of result, randomly select mark result verification as follows:

By date comprision in table it can be seen that Derivation image coordinate differs only small with real image coordinate, illustrating that this scaling method has good validity and reliability, calibration result precision is significantly high. In actual applications, especially, when requiring higher to stated accuracy, the scaling method applying this invention is well positioned to meet the demand to actual stated accuracy, and the method realizes simple, calibration result is stably quick, has very big advantage and using value.

Although present disclosure has been made to be discussed in detail already by above preferred embodiment, but it should be appreciated that the description above is not considered as limitation of the present invention. After those skilled in the art have read foregoing, multiple amendment and replacement for the present invention all will be apparent from. Therefore, protection scope of the present invention should be limited to the appended claims.

Claims (10)

1. a camera marking method, it is characterised in that comprise procedure below:

Determine video camera geometry imaging relations, and introduce distortion of camera correction, obtain introducing the nonlinear imaging model of distortion, and then determine all camera parameters to be solved;

Gather the scaling board image of several different visual angles as reference substance, extract image coordinate information therein, and with camera parameters to be solved for object, substitution differential evolution particle cluster algorithm carries out optimizing solve, to calibrate all of camera parameters, and then realize the three-dimensional space position mapping to two dimensional image, complete the mark of two dimensional image respective coordinates.

2. camera marking method as claimed in claim 1, it is characterised in that definition has:

The first image coordinate system in units of pixel, its initial point o_iIt is positioned at the upper left corner of ccd image plane, u axle and v axle and is respectively parallel to pixel column and the pixel column of the plane of delineation; (u, v) represents object point P pixel coordinate under described first image coordinate system;

With the second image coordinate system that physical unit represents, its initial point o is positioned at the intersection point of camera optical axis and ccd image plane, and x-axis and y-axis are respectively parallel to u axle and v axle; (x_u,y_u) represent object point P image coordinate under described second image coordinate system in desirable pin-hole model; (x_d,y_d) represent object point P real image coordinate under described second image coordinate system in nonlinear imaging model;

Camera coordinate system, its initial point o_cIt is positioned at the photocentre of camera lens, x_cAxle, y_cAxle is respectively parallel to x-axis and y-axis, z_cAxle is the video camera primary optical axis vertical with the plane of delineation; o_cO is the effective focal length f of video camera, (x_c,y_c,z_c) represent object point P coordinate under camera coordinate system;

The world coordinate system of performance three-dimensional space position, (x_w,y_w,z_w) represent object point P world coordinates under world coordinate system;

Further, in described nonlinear imaging model, there is following corresponding relation:

$\left[\begin{array}{c}{x}_u\\ y_u\end{array}\right]=[1+k_1(x_d^2+y_d^2)+k_2{(x_d^2+y_d^2)}^2+k_3{(x_d^2+y_d^2)}^3]\left[\begin{array}{c}{x}_d\\ y_d\end{array}\right]+\left[\begin{array}{c}2p_1{x}_d{y}_d+p_2({3x}_d^2+y_d^2)\\ p_1(x_d^2+3y_d^2)+{2p}_2{x}_d{y}_d\end{array}\right]$

Wherein, k₁,k₂And k₃It is the coefficient of radial distortion needing to demarcate, p₁And p₂It it is the tangential distortion coefficient needing to demarcate.

3. camera marking method as claimed in claim 2, it is characterised in that

By changing the horizontally or vertically orientation angle of scaling board image, gather the scaling board image of several different visual angles; Described scaling board image is the unequal gridiron pattern of ranks dimension, and each fixed point on described scaling board image is each angle point on this gridiron pattern.

4. camera marking method as claimed in claim 3, it is characterised in that

If collecting n width scaling board image, when each image has m fixed point, set up object function as follows:

$f_{\mathrm{obj}}=\min \underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}\left|\right|{\hat{p}}_{\mathrm{ij}}-p(M_A,k_1,k_2,k_3,p_1,p_2,R_i,T_i,P_j)\left|\right|$

It it is the pixel coordinate of jth fixed point in the i-th width scaling board image;

p(M_A,k₁,k₂,k₃,p₁,p₂,R_i,T_i,P_j) it is the world coordinates of jth fixed point, P in the i-th width image_jIt is require over to demarcate the pixel coordinate solved;

M_AFor intrinsic parameters of the camera battle array (f_x,f_y,c_x,c_y), wherein f_x=f/d_x, f_y=f/d_yBeing the scale factor of camera image plane horizontal direction x-axis and vertical direction y-axis respectively, f is the effective focal length of video camera, d_xAnd d_yBe respectively in image plane adjacent two picture points in the horizontal direction with the physical distance in vertical direction, c_xAnd c_yCamera lens center respectively with optical axis in the horizontal direction with the side-play amount of vertical direction;

R_i, T_iIt is the outer parameter that the i-th width image is corresponding, i.e. spin matrix and translation matrix respectively; Wherein, spin matrix R_iComprise three vectors that axial deflection angle tolerance each with three dimensions is corresponding respectively; Translation matrix T_iRepresent the side-play amount between camera coordinate system and world coordinate system.

5. camera marking method as claimed in claim 4, it is characterised in that

Based on described differential evolution particle cluster algorithm described object function is optimized and solves, obtain the optimum stable solution of camera parameters; This algorithm application calibration process is as follows:

Step 1, initialization population: in allowed band, randomly generate position and the speed of N number of particle, and set the bound of particle rapidity;

Step 2, choose fitness function: with the distance of the pixel solved and actual pixels point for target, as fitness evaluation standard, calculate adaptive value and obtain initializing individual extreme value and the global extremum of colony;

The position of step 3, more new particle and speed: be updated according to the iterative formula speed to each particle and position;

Step 4, selection renewal: adopt the selection strategy of differential evolution, compare the adaptive value of the adaptive value of particle after renewal with the front particle of renewal, and the particle selecting fitness high updates position;

Step 5, the operation that intersects: increase the multiformity of colony, it is ensured that the individuality that fitness is high;

Step 6, mutation operation: by comparing fitness, make low fitness individuality make a variation, and forms colony of a new generation;

Step 7, based on described iterative formula, and the adaptive value according to colony of a new generation, update individual extreme value and the global extremum of colony;

Step 8, judge whether to meet end condition: if reaching maximum iteration time, then end loop output result, otherwise forwards step 3 to and continues executing with.

6. camera marking method as claimed in claim 5, it is characterised in that

Fitness function definition in step 2 is as follows:

$f\left(\right)=\min \underset{i=1}{\overset{m}{\mathrm{\Σ}}}\sqrt{{(u_i-x_i)}^2+{(v_i-y_i)}^2}$

This fitness function is the distance calculating required pixel and actual pixels point, and distance is more little represents that individual fitness is more high.

7. camera marking method as claimed in claim 6, it is characterised in that

Iterative formula in step 3 is:

$v_{\mathrm{id}}^{k+1}={\mathrm{wv}}_{\mathrm{id}}^k+c_1\mathrm{rand}\left(\right)(\mathrm{pbest}-x_{\mathrm{id}}^k)+c_2\mathrm{rand}\left(\right)(\mathrm{gbest}-x_{\mathrm{id}}^k)$

$x_{\mathrm{id}}^{k+1}=x_{\mathrm{id}}^k+v_{\mathrm{id}}^{k+1}$

If the information D dimensional vector of particle i represents, wherein positional representation is X_i=(x_i1,x_i2,...,x_iD)^T, speed is V_i=(v_i1,v_i2,...,v_iD)^T; W is inertia weight; Rand () is the random number between [0,1]; Pbest is the particle i position at the d individual extreme point tieed up; Gbest is the whole population position at the d global extremum point tieed up; c₁, c₂It is accelerated factor, is respectively regulated to the maximum step-length of the overall situation preferably particle and the flight of individuality best particle direction; K is iterations.

8. camera marking method as claimed in claim 7, it is characterised in that

The selection operation operator used in step 4 is as follows:

9. camera marking method as claimed in claim 8, it is characterised in that

The crossover operator of operation of intersecting in step 5 is:

Wherein, j=1,2 ..., N, j_rand�� [1, N]; CR is cross-over control parameter.

10. camera marking method as claimed in claim 9, it is characterised in that

In step 6, the mutation operator of mutation operation computing is:

v_i,G=x_r1,G+F(x_r2,G-x_r3,G)

Wherein, r1, r2, r3 is the integer being arbitrarily not mutually equal between [1, N]; F is variance control parameter.