CN102867304B - Method for establishing relation between scene stereoscopic depth and vision difference in binocular stereoscopic vision system - Google Patents

Abstract

The invention discloses a method for establishing a relation between a scene stereoscopic depth and a vision difference in a binocular stereoscopic vision system. the method comprises the steps of firstly, solving inner parameters, and relative rotary matrixes and translation vectors of left and right cameras; then, analyzing a main error source and an error model of the binocular stereoscopic vision system; then, analyzing influences of the main error on a base line length and the vision difference of the parallel binocular stereoscopic vision system; then, establishing a common relation model of the scene stereoscopic depth and the vision difference in the binocular stereoscopic vision system; obtaining depth information by a laser distance measuring instrument by selecting a certain amount of demarcation points and carrying out demarcation based on a least square method to solve a relation model between the scene stereoscopic depth and the vision difference in the binocular stereoscopic vision system; and finally, solving the vision difference in left and right images through a corresponding fixed matching method so as to realize accurate recovery and three-dimensional reestablishment of the scene stereoscopic depth. The method has the beneficial effect of directly, simply, accurately and extremely improving the accuracy of the stereoscopic depth recovery and the three-dimensional reestablishment.

Description

The relation establishing method of Binocular Stereo Vision System Scene three-dimensional depth and parallax

Technical field

The scene depth that the present invention relates to a kind of stereo visual system of arbitrary disposition recovers and three-dimensional rebuilding method, be specifically related to the model of the universal relation of a kind of run-in index Binocular Stereo Vision System Scene three-dimensional depth and parallax, be applicable to machine vision and photogrammetric in scene three-dimensional depth recover and three-dimensional reconstruction.Belong to advanced to manufacture and automatic field.

Background technology

The research of binocular stereo vision is one of hot issue of field of machine vision, and the three-dimensional depth utilizing Binocular Stereo Vision System can realize scene point recovers and three-dimensional reconstruction.The three-dimensional depth of scene recovers to be obtain scene point relative to certain video camera along the axial distance of key light according to the pixel coordinate of the corresponding diagram picture point of scene point in stereo visual system.And three-dimensional reconstruction is exactly on the basis of depth recovery, obtain the three-dimensional coordinate of scene point under a certain camera coordinate system further.

Binocular stereo vision is generally divided into convergence type and run-in index, and wherein run-in index binocular stereo vision is a kind of special case of convergence type binocular tri-dimensional vision model.Run-in index Binocular Stereo Vision System middle left and right two camera properties of standard are identical, rotation relationship is there is not between camera coordinate system, namely the optical axis of two video cameras needs completely parallel, and only there is translation relation along certain direction in left and right two camera coordinate systems, translational movement is commonly referred to base length, for the run-in index Binocular Stereo Vision System of standard, the relational model of scene three-dimensional depth and parallax is known by stereoscopic vision field.But under real world conditions, the idealized configuration of the run-in index Binocular Stereo Vision System of standard is difficult to realize, and its main cause has two aspects: on the one hand, and the inner parameter of left and right two video cameras can not reach incomplete same; On the other hand, it is impossible for by mechanical means, the position relationship of left and right two video cameras being adjusted to complete parallel placement.For off-gauge run-in index Binocular Stereo Vision System, lens distortion, perspective error and systematic error equal error source is mainly there is in Binocular Stereo Vision System, cause the scene three-dimensional depth of Binocular Stereo Vision System and parallax relational model to there occurs certain change, the degree of accuracy of the recovery of scene three-dimensional depth and three-dimensional reconstruction is decreased.

At present, directly scene depth recovery and three-dimensional reconstruction can be carried out owing to also there is no a kind of method, so existing method is all adopt round-about way to carry out scene depth to recover and three-dimensional reconstruction, mainly be divided into two classes: the first kind is the image be converted into by the image in unparallel shaft stereo visual system by the method for image rectification in standard parallel axle stereo visual system, then calls depth recovery and the three-dimensional rebuilding method of standard parallel axle stereo visual system; Another Equations of The Second Kind is by the projection line reverse extending of scene point in two video cameras, is then found in scene by optimization method and nearestly with two projection lines a bit carries out depth recovery and three-dimensional reconstruction.Equations of The Second Kind method needs first to carry out initial estimate usually, then carries out iteration optimizing, not only time-consuming but also easily converge to locally optimal solution, thus seldom adopts in actual applications.So, in stereoscopic vision, carry out at present scene depth to recover and three-dimensional reconstruction generally adopts first kind method, but need first to carry out image rectification when adopting first kind method, this is also a time-consuming step, and error can be produced in the conversion process of image rectification, affect scene three-dimensional depth and parallax relational model, make the degree of accuracy of the recovery of scene three-dimensional depth and three-dimensional reconstruction reduce equally.Therefore, this area needs are a kind of is applicable to the scene three-dimensional depth of the run-in index Binocular Stereo Vision System of arbitrary disposition and the universal model of parallax.

Summary of the invention

For solving the problem, the main error that the object of the invention is to exist for run-in index Binocular Stereo Vision System is on the impact of scene three-dimensional depth and parallax relational model, the universal relation model of a kind of run-in index Binocular Stereo Vision System Scene three-dimensional depth and parallax is proposed, to realize accurate recovery and the three-dimensional reconstruction of scene three-dimensional depth, and neither need the image of Binocular Stereo Vision System carrying out polar curve correction, do not need to carry out initial estimate and iteration optimizing, have directly yet, simply, advantage accurately.

For solving the problem, the present invention takes following technical scheme to realize:

A relation establishing method for Binocular Stereo Vision System Scene three-dimensional depth and parallax, is characterized in that, comprises the following steps:

1) to be demarcated by monocular-camera and the method for Binocular Stereo Vision System demarcation, obtain the intrinsic parameter of left and right cameras and relative rotation matrices and translation vector;

2) by the pin-hole model of video camera perspective imaging, according to the change of Binocular Stereo Vision System base length and parallax, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax is set up;

3) by choosing the calibration point of some, obtaining depth information by laser range finder, carrying out the demarcation based on least square method, obtain the relational model of given Binocular Stereo Vision System scene three-dimensional depth and parallax;

4) for left and right cameras obtain two width images of Same Scene, by the method for corresponding point matching, obtain the scene parallax in left images;

5) by scene parallax, Exact recovery and the three-dimensional reconstruction of scene three-dimensional depth is realized.

The relation establishing method of aforesaid Binocular Stereo Vision System Scene three-dimensional depth and parallax, is characterized in that, in described step 2) in, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model step of parallax are:

21) only considering in model error situation, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax are such as formula (18):

$z=a+\frac{b}{d}---\left(18\right)$

Wherein, a and b is constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain;

22) under considering model error and matching error situation at the same time, completing steps 21) after, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax are formula (20)

$z=a+\frac{b}{d+\mathrm{\Δd}}=\frac{\mathrm{ad}+\mathrm{a\Δd}+b}{d+\mathrm{\Δd}}=\frac{A^{\′}d+B^{\′}}{d+C^{\′}}---\left(20\right)$

Wherein, A ', B ' and C ' are constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain, and z represents scene depth, d is the parallax of scene point in left and right cameras image, and the converted quantity of parallax is Δ d.

Utilize the relational implementation three-dimensional depth of the aforesaid Binocular Stereo Vision System neutral body degree of depth and parallax to recover and the method for three-dimensional reconstruction, it is characterized in that, comprise the following steps:

(1) to be demarcated by monocular-camera and the method for Binocular Stereo Vision System demarcation, obtain the intrinsic parameter of left and right cameras and relative rotation matrices and translation vector;

(2) by the pin-hole model of video camera perspective imaging, according to the change of run-in index Binocular Stereo Vision System base length and parallax, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax is set up;

(3) two width images of scaling board in the Same Scene obtained for left and right cameras, by the method for Point matching, are extracted corresponding match point, obtain the pixel coordinate of corresponding point, and asked for the parallax of corresponding point by Euclidean distance.

(4) laser range finder is demarcated, make laser range finder and calibrating template keep vertical, ensure that camera coordinate system and laser range finder are in the same coordinate system simultaneously, thus under obtaining current parallax, the degree of depth between video camera to scaling board;

(5) mobile optical translation stage, repeats the operation of (3)-(4), and obtain the data of many group parallaxes and the degree of depth, in moving process, scaling board keeps vertical with video camera and laser range finder.

(6) carry out least square fitting by the parallax of corresponding point and the depth value of scene point in camera coordinate system as a pair data, thus obtain corresponding parameter, obtain the scene three-dimensional depth of given Binocular Stereo Vision System and the relational model of parallax;

(7) for left and right cameras obtain two width images of Same Scene, the method for surely being mated by correspondence, obtains the scene parallax in left images;

(8) by scene parallax, Exact recovery and the three-dimensional reconstruction of scene three-dimensional depth is completed.

Usefulness of the present invention is, the present invention considers the impact on scene three-dimensional depth and parallax relational model of the main error that exists in Binocular Stereo Vision System, establish and be a kind ofly applicable to the scene three-dimensional depth of the run-in index Binocular Stereo Vision System of arbitrary disposition and the universal model of parallax, and neither need the image of Binocular Stereo Vision System carrying out polar curve correction, do not need to carry out initial estimate and iteration optimizing yet, have directly, simply, advantage accurately, drastically increase three-dimensional depth and recover and the degree of accuracy of three-dimensional reconstruction.

Accompanying drawing explanation

Fig. 1 is standard parallel formula Binocular Stereo Vision System geometric model;

Fig. 2 is camera lens distortion schematic diagram;

Fig. 3 is Binocular Stereo Vision System error separation;

Fig. 4 is binocular stereo vision parallax and Depth Information Acquistion installation drawing;

Fig. 5 is the parameter acquisition procedure figure of binocular stereo vision universal relation model.

Embodiment

Below in conjunction with accompanying drawing, concrete introduction is done to the present invention:

Fig. 1 is the geometric model of standard parallel formula Binocular Stereo Vision System.As shown in Figure 1, standard parallel formula Binocular Stereo Vision System is made up of two identical video cameras, and two planes of delineation are positioned in a plane, and the coordinate axis of two video cameras is parallel to each other, and x-axis overlaps.If O _l, O _rfor the optical centre position of left and right cameras, O _lwith O _rbetween distance be baseline B, the focal length of video camera is f, the scene point p subpoint in the left and right plane of delineation is p _land p _r.The ideal relationship model of standard parallel formula Binocular Stereo Vision System scene three-dimensional depth and parallax is such as formula (1):

$z=\frac{\mathrm{fB}}{d}---\left(1\right)$

Wherein, d is scene parallax, and z represents scene depth.

In practical situations both, in camera optics imaging process, camera lens also exists radial distortion, tangential distortion and thin prism distortion, causes image coordinate to there is distortion, so Binocular Stereo Vision System introduces model error; The Binocular Stereo Vision System simultaneously adopted not is the run-in index Binocular Stereo Vision System of standard, incomplete same and the position relationship of left and right cameras of the intrinsic parameter of left and right cameras can not complete parallel placement, causes Binocular Stereo Vision System to introduce perspective error, polar curve correction error, systematic error etc.

Fig. 2 is camera lens distortion schematic diagram.As shown in Figure 2, radial distortion in video camera imaging process, tangential distortion and thin prism distortion introduce radial distortion, and tangential distortion and thin prism distortion introduce tangential distortion, and the error that radial distortion and tangential distortion system are introduced is referred to as model error.Image physical coordinates during scene point consideration model error is such as formula (2):

$x_d=x_u+{\mathrm{\δ}}_{x_u}(x_u,y_u)$

(2)

$y_d=y_u+{\mathrm{\δ}}_{y_u}(x_u,y_u)$

Wherein, (x _u, y _u) be the ideal position image physical coordinates of scene point, (x _d, y _d) be the location drawing after scene point distortion as physical coordinates, be the impact of model error on the image physical coordinates of scene point, specifically can be expressed as formula (3):

${\mathrm{\δ}}_{x_u}=(k_1{R}^2+k_2{R}^4+k_5{R}^6)x_u+2k_3{x}_u{y}_u+k_4(R^2+{{2x}_u}^2)$

(3)

${\mathrm{\δ}}_{y_u}=(k_1{R}^2+k_2{R}^4+k_5{R}^6)y_u+k_3(R^2+2{y_u}^2)+{2k}_4{x}_u{y}_u$

R ²=x _u ²+ y _u ², k ₁, k ₂, k ₅for camera lens radial distortion parameter, k ₃, k ₄for tangential distortion parameter.

For general video camera, above-mentioned radial distortion can describe the nonlinear distortion of camera lens, and too much nonlinear parameter can reduce stability of solution on the contrary, and therefore emphasis of the present invention considers the radial distortion of camera lens, namely

${\mathrm{\δ}}_{x_u}=(k_1{R}^2+k_2{R}^4+k_5{R}^6)x_u$

(4)

${\mathrm{\δ}}_{y_u}=(k_1{R}^2+k_2{R}^4{k}_5{R}^6)y_u$

Then single camera model error can be expressed as formula (5):

$\mathrm{\Δ\δ}=\sqrt{{(x_d-x_u)}^2+{(y_d-y_u)}^2}$

$=\sqrt{{{\mathrm{\δ}}_{x_u}}^2+{{\mathrm{\δ}}_{y_u}}^2}---\left(5\right)$

$=\sqrt{{(k_1{R}^2+k_2{R}^4+k_5{R}^6)}^2{x_u}^2+{(k_1{R}^2+k_2{R}^4+k_5{R}^6)}^2{y_u}^2}$

The model error of video camera can be reduced by some nonlinear calibration methods.

When utilizing Binocular Stereo Vision System to realize the three-dimensional reconstruction of scene point, that usually choose Minimum Mean Square Error and that (SSD) mates as scene point similarity measure.In the ideal case in match window (window of 9*9 pixel size) mean square deviation and be expressed as formula (6):

$M\left(d\right)=\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}{(I_l(x_u,y_u)-I_r(x_u-d,y_u))}^2---\left(6\right)$

Wherein, (x _u, y _u) be the ideal position image physical coordinates of scene point, d is parallax, I _l, I _rthe gray scale at sight spot, left and right cameras image midfield respectively.

In practical situations both, the optical axis non-critical of left and right cameras is parallel, namely there is certain anglec of rotation, the picture of scene point in the left and right cameras plane of delineation not only also exists translation and there is certain rotational deformation simultaneously, Binocular Stereo Vision System introduces perspective distortion error, and the gray scale of the right camera review scene point after therefore introducing perspective distortion error can be expressed as formula (7):

I _r(x _f,y _f)=I _l(x _u+d ₀+ay _u,y _u) （7）

Wherein, a is the non-zero gradient of disparity on y-axis direction.

Formula (7) is launched, such as formula (8) by Taylor's formula:

$I_l(x_u+d_0+{\mathrm{ay}}_u,y_u)\≈I_l(x_u,y_u)+(d_0+{\mathrm{ay}}_u)\frac{{\∂I}_l}{{\∂x}_u}---\left(8\right)$

Formula (8) is brought into formula (6), obtain the mean square deviation under actual conditions and as follows:

$M\left(d\right)\≈\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}{\left((d-d_0-{\mathrm{ay}}_u)\frac{{\∂I}_l}{{\∂x}_u}\right)}^2---\left(9\right)$

Work as d=d ₀+ ae _ftime, mean square deviation with minimum.Therefore, perspective distortion error can represent such as formula (10):

$e_f=\frac{\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}{y}_u{\left|\frac{{\∂I}_l}{{\∂x}_u}\right|}^2}{\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}{\left|\frac{{\∂I}_l}{{\∂x}_u}\right|}^2}---\left(10\right)$

Scene depth for off-gauge run-in index Binocular Stereo Vision System recovers and three-dimensional reconstruction, main method is corrected by the image in off-gauge run-in index Binocular Stereo Vision System to the image pair of run-in index Binocular Stereo Vision System being converted into standard by the polar curve of stereoscopic image, and also can produce error in the polar curve trimming process that image is right.Suppose that polar curve corrects on rear y-axis direction and there is very little misalignment degree m, now the gray scale of right camera review scene point can be expressed as formula (11):

I _r(x _m,y _m)=I _l(x _u+d ₀,y _u+m) （11）

In like manner, formula (11) is launched, such as formula (12) by Taylor's formula:

$I_l(x_u+d_0,y_u+m)\≈I_l(x_u,y_u)+d_0\frac{{\∂I}_l}{{\∂x}_u}+m\frac{{\∂I}_l}{{\∂y}_u}---\left(12\right)$

Formula (12) is brought into formula (6), obtain the mean square deviation under actual conditions and as follows:

$M\left(d\right)\≈\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}{((d-d_0)-\frac{{\∂I}_l}{{\∂x}_u}-m\frac{{\∂I}_l}{{\∂y}_u})}^2---\left(13\right)$

Work as d=d ₀+ me _mtime, mean square deviation with minimum.Therefore, polar curve correction error can be expressed as formula (14):

$e_m=\frac{\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}\frac{{\∂I}_l}{{\∂x}_u}\·\frac{{\∂I}_l}{{\∂y}_u}}{\underset{x=-4}{\overset{4}{\mathrm{\Σ}}}\underset{y=-4}{\overset{4}{\mathrm{\Σ}}}{\left|\frac{{\∂I}_l}{{\∂x}_u}\right|}^2}---\left(14\right)$

Meanwhile, run-in index Binocular Stereo Vision System also exists systematic error, and systematic error mainly comprises window effect error and linear error.Perspective error, polar curve correction error and systematic error are referred to as matching error.

Fig. 3 is run-in index Binocular Stereo Vision System error separation.As shown in Figure 3, O _l, O _rrepresent the photocentre of left and right cameras, O _lz _land O _rz _rbe respectively the optical axis of left and right video camera, two optical axises are parallel, O _land O _rbetween distance be baseline B, the focal length of video camera is f.

Set up two-dimensional coordinate system O _ro _lz, with O _lfor true origin, O _lo _rfor X-axis, O _lz is Z axis, then O _l(0,0), O _r(B, 0), as plane on plane Z=f.If p ₁, p ₂for calibration point, conveniently derive, if its coordinate is respectively (x ₁, z), (x ₂, z), then put p ₁and p ₂ideal position in left and right cameras coordinate system is respectively p _l1, p _r1and p _l2, p _r2.Because left and right cameras inner parameter is incomplete same, therefore suppose that scene point is respectively Δ δ at the model error of left and right cameras ₁with Δ δ ₂.

Consider model error, p ₁and p ₂physical location in left and right cameras is respectively p _l1', p _r1' and p _l2', p _r2', the photocentre position of the video camera that can calculate accordingly is O _l', O _r', some p _l1coordinate be (fx ₁/ z, f), some p _l2coordinate be (fx ₂/ z, f), some p _l1' coordinate (fx ₁/ z+ Δ δ ₁, f), some p _l2' coordinate (fx ₂/ z+ Δ δ ₁, f), some p _r1coordinate be (B+f (x ₁-B)/z, f), some p _r2coordinate be (B+f (x ₂-B)/z, f), some p _r1' coordinate be (B+f (x ₁-B)/z+ Δ δ ₂, f), some p _r2' coordinate be (B+f (x ₂-B)/z+ Δ δ ₂, f), by p ₁p _l1' and p ₂p _l2' can O be obtained _l' coordinate (z Δ δ ₁/ (z-f), 0), in like manner obtain O _r' coordinate be (B+z Δ δ ₂/ (z-f), 0), now the base length of stereo visual system is:

B′=B+z(Δδ ₂-Δδ ₁)/(z-f)＝B+ΔB （15）

The introducing of model error makes base length there occurs change.

If p ₃for test point, p ₃the coordinate of point is (x ₃, z), then p ₃picture point in left and right cameras plane is p _l3' and p _r3', consider model error Δ δ and matching error Δ ε, the p obtained simultaneously ₃picture point is p _l3' and p _r3', obtain a p according to linear model ₃'.P _l3' coordinate be (fx ₃/ z+ Δ δ ₁, f), p _r3' coordinate be (B+f (x ₃-B)/z+ Δ δ ₂, f), p _r3" coordinate of point is (B+f (x ₃-B)/z+ Δ δ ₂+ Δ ε, f), now p ₃the parallax of point is:

d′=(z-f)B/z+Δε+Δδ ₂-Δδ ₁=d+Δd （16）

When only considering model error, the base length of run-in index Binocular Stereo Vision System there occurs change, if the variable quantity of base length is Δ B, the depth error that the change of base length causes is Δ z, then scene point three-dimensional depth and parallax relational model can be expressed as:

$z=\frac{f(B+z({\mathrm{\Δ\δ}}_2-{\mathrm{\Δ\δ}}_1)/(z-f))}{d}+\mathrm{\Δz}=\frac{f(B+\mathrm{\ΔB})}{d}+\mathrm{\Δz}---\left(17\right)$

Wherein, f is the focal length of video camera, and d is the parallax of scene point in left and right cameras image.

When the base length information of stereo visual system is inaccurate or there are some unknown errors, formula (17) still also exists larger error, and therefore formula (17) is extended to general type, such as formula (18) by the present invention:

$z=a+\frac{b}{d}---\left(18\right)$

Wherein, a and b is constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain.

When to consider model error and matching error simultaneously, base length and the parallax information of run-in index Binocular Stereo Vision System all there occurs conversion, if the variable quantity of base length is Δ B, the converted quantity of parallax is Δ d, the depth error that the change of base length causes is Δ z, formula (17) is promoted, such as formula (19) by the present invention further:

$z=\frac{f(B+\mathrm{\ΔB})}{d+\mathrm{\Δd}}+\mathrm{\Δz}=\frac{\mathrm{d\Δz}+f(B+\mathrm{\ΔB})+\mathrm{\Δd\Δz}}{d+\mathrm{\Δd}}---\left(19\right)$

Formula (18) is promoted, such as formula (20):

$z=a+\frac{b}{d+\mathrm{\Δd}}=\frac{\mathrm{ad}+\mathrm{a\Δd}+b}{d+\mathrm{\Δd}}=\frac{A^{\′}d+B^{\′}}{d+C^{\′}}---\left(20\right)$

Wherein, A ', B ' and C ' are constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain.

For left and right cameras obtain two width images of Same Scene, the method for surely being mated by correspondence, obtains the scene parallax in left images.By scene parallax, complete Exact recovery and the three-dimensional reconstruction of scene three-dimensional depth.

Fig. 4 is binocular stereo vision parallax and Depth Information Acquistion installation drawing, and this device is made up of optical translation platform (containing scaling board), image capture device, laser range finder and integrated information processing platform four part; Optical translation platform is made up of calibrating template and single shaft mobile platform, and template is fixed on the platform that horizontal direction moves freely, thus realizes the translation in horizontal direction.Image capture device is made up of arbitrary binocular stereo vision camera, realizes the collection that image is right.Laser range finder adopts one dimension laser range finder, and measuring accuracy can reach 2mm.The information processing platform comprises PC and corresponding program, carries out treatment and analysis to the stereo pairs obtained and range information.

Fig. 5 is the parameter acquisition procedure figure of binocular stereo vision universal relation model, concrete steps as shown in the figure:

(1) scaling board is placed in the public view field of two video cameras, uses binocular camera to carry out image acquisition to above-mentioned scaling board simultaneously and obtain image pair

(2) two width images of scaling board in the Same Scene obtained for left and right cameras, by the method for corresponding point matching, extract corresponding match point.

(3) for a pair match point in left images, obtain the pixel coordinate of corresponding point, and asked for the parallax of corresponding point by Euclidean distance.

(4) laser range finder is demarcated, make laser range finder and calibrating template keep vertical, ensure that camera coordinate system and laser range finder are in the same coordinate system simultaneously, thus under obtaining current parallax, the degree of depth between video camera to scaling board.

(5) mobile optical translation stage, ensures in moving process, and scaling board keeps vertical with video camera and laser range finder, thus obtains the data of many group parallaxes and the degree of depth.

(6) carry out least square fitting by the parallax of corresponding point and the depth value of scene point in camera coordinate system as one group of data, thus obtain corresponding parameter and obtain the scene three-dimensional depth of given Binocular Stereo Vision System and the relational model of parallax;

Exact recovery and the three-dimensional reconstruction of scene three-dimensional depth can be realized by above method and step.

It should be noted that, above-described embodiment does not limit the present invention in any form, the technical scheme that the form that all employings are equal to replacement or equivalent transformation obtains, and all drops within protection scope of the present invention.

Claims (2)

1. a relation establishing method for Binocular Stereo Vision System Scene three-dimensional depth and parallax, is characterized in that, comprises the following steps:

1) to be demarcated by monocular-camera and the method for Binocular Stereo Vision System demarcation, obtain the intrinsic parameter of left and right cameras and relative rotation matrices and translation vector;

2) by the pin-hole model of video camera perspective imaging, according to the change of Binocular Stereo Vision System base length and parallax, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax is set up;

The universal relation model step of the scene three-dimensional depth and parallax of setting up Binocular Stereo Vision System is:

21) only considering in model error situation, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax are such as formula (18):

$z=a+\frac{b}{d}---\left(18\right)$

Wherein, a and b is constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain;

22) under considering model error and matching error situation at the same time, completing steps 21) after, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax are formula (20)

$z=a+\frac{b}{d+\mathrm{\Δd}}=\frac{\mathrm{ad}+\mathrm{a\Δd}+b}{d+\mathrm{\Δd}}=\frac{A^{\′}d+B^{\′}}{d+C^{\′}}---\left(20\right)$

Wherein, A ', B ' and C ' are constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain, and z represents scene depth, d is the parallax of scene point in left and right cameras image, and the converted quantity of parallax is Δ d;

3) by choosing the calibration point of some, obtaining depth information by laser range finder, carrying out the demarcation based on least square method, obtain the relational model of given Binocular Stereo Vision System scene three-dimensional depth and parallax;

4) for left and right cameras obtain two width images of Same Scene, by the method for corresponding point matching, obtain the scene parallax in left images;

5) by scene parallax, Exact recovery and the three-dimensional reconstruction of scene three-dimensional depth is realized.

2. utilize the relational implementation three-dimensional depth of the Binocular Stereo Vision System neutral body degree of depth and parallax to recover and the method for three-dimensional reconstruction, it is characterized in that, comprise the following steps:

(1) to be demarcated by monocular-camera and the method for Binocular Stereo Vision System demarcation, obtain the intrinsic parameter of left and right cameras and relative rotation matrices and translation vector;

(2) by the pin-hole model of video camera perspective imaging, according to the change of run-in index Binocular Stereo Vision System base length and parallax, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax is set up;

The universal relation model step of the scene three-dimensional depth and parallax of setting up Binocular Stereo Vision System is:

21) only considering in model error situation, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax are such as formula (18):

$z=a+\frac{b}{d}---\left(18\right)$ Wherein, a and b is constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain;

22) under considering model error and matching error situation at the same time, completing steps 21) after, the scene three-dimensional depth of Binocular Stereo Vision System and the universal relation model of parallax are formula (20)

$z=a+\frac{b}{d+\mathrm{\Δd}}=\frac{\mathrm{ad}+\mathrm{a\Δd}+b}{d+\mathrm{\Δd}}=\frac{A^{\′}d+B^{\′}}{d+C^{\′}}---\left(20\right)$

Wherein, A ', B ' and C ' are constant, and the parallax information of its depth information and correspondence thereof of solving by gathering some calibration points utilizes least square method to obtain, and z represents scene depth, d is the parallax of scene point in left and right cameras image, and the converted quantity of parallax is Δ d;

(3) two width images of scaling board in the Same Scene obtained for left and right cameras, by the method for Point matching, are extracted corresponding match point, obtain the pixel coordinate of corresponding point, and asked for the parallax of corresponding point by Euclidean distance;

(4) laser range finder is demarcated, make laser range finder and calibrating template keep vertical, ensure that camera coordinate system and laser range finder are in the same coordinate system simultaneously, thus under obtaining current parallax, the degree of depth between video camera to scaling board;

(5) mobile optical translation stage, repeats the operation of (3)-(4), and obtain the data of many group parallaxes and the degree of depth, in moving process, scaling board keeps vertical with video camera and laser range finder;

(6) least square fitting is carried out by the parallax of corresponding point and the depth value of scene point in camera coordinate system as a pair data, thus obtain corresponding parameter, obtain the scene three-dimensional depth of given Binocular Stereo Vision System and the relational model of parallax;

(7) for left and right cameras obtain two width images of Same Scene, by the method for corresponding point matching, obtain the scene parallax in left images;

(8) by scene parallax, Exact recovery and the three-dimensional reconstruction of scene three-dimensional depth is completed.