CN102622751A - Image processing method of three-dimensional camera - Google Patents

Abstract

The invention belongs to the image data and processing technology, and particularly relates to an image processing method enabling shot images to be converted from two dimensions into three dimensions and applied to a 3D (three dimensions) camera. An active visual platform is used for controlling the camera to perform a group of three-orthogonal translational motions, and plane structure information can be recovered and Euclidean reconstruction is performed to recover three-dimensional image information through image corresponding points and translational motion distance. The image processing method does not need calibration objects, but needs to control the camera to perform a specific motion. Inner parameters of the camera can be calculated by using specific characteristics of the motion.

Description

A kind of image processing method of three-dimensional camera

Technical field

The invention belongs to view data and treatment technology, relate more specifically in a kind of 3D of being applied to (Three Dimensions) camera photographic images by the disposal route of two-dimensional transformations to 3-D view.

Background technology

3D is meant three-dimensional, three dimensions, three coordinates, promptly have long, have wide, height is arranged, in other words three-dimensional, be for having only long and wide plane.

We just lived in the three-dimensional solid space originally; This world that our eyes and health perceive all is 3 D stereo; And have abundant colors, gloss, surface, material or the like appearance tactile impression, and the kinematic relation of ingenious and complicated inner structure and space-time dynamic; We are three-dimensional to any discovery in this world and the original impulsion of creation.But in the very long historical progress of the mankind; Office is in the restriction of technical conditions; Can't use three-dimensional mode directly perceived to describe this three-dimensional world easy, directly, quickly, can only on the two dimensional surface of sandy soil, sheepskin, paper, express and transmit understanding and creation this world with image; People have invented methods such as plane projection and perspective, and have formed abstract 2D plane civilization system based on the paper plane.

The invention of computer with popularize fast, follow the extension at full speed of internet, promptly change and reduced everything, profoundly rewriting our life style, consumption pattern, working method and mode of production.Three-dimensional digital technology based on computer and internet makes people the understanding in the real three-dimensional world revert to again the boundary of original solid directly perceived finally.No matter on the virtual network still in the life in reality; From greatly to aircraft, steamer, automobile, power station, mansion, building, bridge; Little to the life in each slight industrial products; Can both see digitized 3D model, animation and the emulation of computer photograph everywhere, this is the deep revolutions of 2D plane epoch to the 3D digital times.

Open the 3D based on image principle, oversimplify its principle, also very simple.We can do an experiment: two hands are taken pen or chopsticks simultaneously, close eyes, only use the another eyes, attempt the pen in two hands or chopsticks point to together.You can find to accomplish the difficulty of this action than the imagination.Eyes see that object is a two dimensional image, utilize relevant size that object provides and visual cues such as overlapping, can judge the front and back ordering that is positioned at these objects before the background, but can't know but between them actually distance how far.People's vision system is based on two eyes; Two horizontal eyes are when seeing same object, because that residing angle has is different slightly, so the image of seeing difference slightly; Here it is so-called parallax; Brain combines this two width of cloth picture, and automatic synthesis analysis just forms a kind of vision of the degree of depth.Simultaneously, brain can also be based in two width of cloth images that receive, and the size of potential difference between the same object is judged the degree of depth and the distance of object, and is far away more apart from eyes, and potential difference is just more little, otherwise just big more.The ultimate principle of Here it is 3D image.

The 3D digital camera is meant the digital camera that can appreciate stereogram picture or animation with bore hole.The birth of 3D digital camera just means that also people can use professional glasses, with the naked eye just can enjoy the effect of stereo-picture.The 3D digital camera generally is equipped with 2 camera lenses, so that can reproduce stereopsis.When we see object with eyes, about catch object from slightly different angle respectively.After the delicate difference of image for this reason, these two images were synthesized in brain, we just can grasp the profile of object three-dimensionally.Therefore, on the theory, two camera lenses just can reproduce stereopsis about camera was loaded onto.But original camera technology can not realize approaching the function of human eye, even two shutters open and close simultaneously about letting, also can produce small time deviation, causes departing from of left and right sides image.

For confirming three-dimensional geometry position and its mutual relationship between the corresponding point in image of certain point of space object surface, must set up the geometric model of camera imaging, these geometric model parameters are exactly camera parameter.These parameters must just can obtain through experiment and calculating under most of conditions, and this process of finding the solution parameter just is referred to as camera calibration (or camera calibration).No matter be in image measurement or machine vision applications, the demarcation of camera parameter all is unusual the key link, and camera calibration is the basis of computer vision research, has important use aspect three-dimensional reconstruction and the target following location.In high-precision visual task, generally all adopt scaling board to carry out calibration of camera internal parameters.The regular reference mark of cloth on the scaling board, their coordinate is known in advance, and positioning error is usually in the μ level.And the problem between monumented point does not obtain enough attention for a long time, generally all adopts manual mode to specify corresponding relation, makes that camera calibration still is a consuming time and process of requiring great effort.

The traditional camera standardization need be demarcated thing all the time in calibration process, and the making precision of demarcation thing can influence calibration result.Some occasion is not suitable for placing the application that the demarcation thing has also limited the traditional camera standardization simultaneously.

Summary of the invention

The camera calibration method based on active vision of the present invention proposes is meant that some movable information of known camera demarcates camera.This method need not demarcated thing, but needs the control camera to do special exercise, utilizes the singularity of this motion can calculate the camera inner parameter.

Algorithm of the present invention only need utilize active vision platform control camera to do three quadrature translation motions; Just can recover planar structure information and carry out the Euclidean reconstruction through the distance of image corresponding point and translation motion then, active vision platform control camera is done three quadrature translation motions and is only carried out one group.

Come the analogue camera imaging process with the pin-hole imaging model, concrete steps are:

$\mathrm{\α}{\left[\begin{array}{ccc}u& v& 1\end{array}\right]}^T=\left[\begin{array}{ccc}f& \mathrm{\ζ}& u_0\\ 0& \mathrm{af}& v_0\\ 0& 0& 1\end{array}\right]\left[\begin{array}{cc}R& t\end{array}\right]{\left[\begin{array}{cccc}X& Y& Z& 1\end{array}\right]}^T$

Wherein α is the chi factor, and R and t are respectively the rotation matrix and the translation vector of corresponding world coordinate system.

Utilize three translation distances of active vision platform measuring and record camera, as the P of camera from beginning ₀, take turns doing the quadrature translation motion to P ₁, P ₂, P ₃, utilize position P ₀As world coordinate system, note P ₀To P ₁, be T ₁, P ₁, to P ₂Be T ₂, P ₂, to P ₃Be T ₃, and satisfy (i=1,2,3, j ≠ i), note P ₁, P ₂, P ₃The position with respect to P ₀Displacement be respectively T ₁₀=T ₁, T ₂₀=T ₁+ T ₂, T ₃₀=T ₁+ T ₂+ T ₃

When three-dimensional point is positioned at plane ∏, there are mapping relations one to one in the image corresponding point, and homography matrix H capable of using representes:

β is a scale factor, H _∞Be the infinite distance unit matrix, N and d are respectively the unit normal vector of plane ∏ and arrive the distance of initial point.

The recovering step of planar structure: N ^TT ₁₀=d λ ₁₀, N ^TT ₂₀=d λ ₂₀, N ^TT ₃₀=d λ ₃₀, translation vector is carried out normalization can get three set of equations as follows: $\frac{N^T{T}_1}{{\mathrm{PT}}_{1}P}=\frac{d{\mathrm{\λ}}_{10}}{P{T}_{1}P},$ $\frac{N^T{T}_2}{{\mathrm{PT}}_{2}P}=\frac{d({\mathrm{\λ}}_{20}-{\mathrm{\λ}}_{10})}{{\mathrm{PT}}_{2}P},$ $\frac{N^T{T}_3}{{\mathrm{PT}}_{3}P}=\frac{d({\mathrm{\λ}}_{30}-{\mathrm{\λ}}_{20}-{\mathrm{\λ}}_{10})}{{\mathrm{PT}}_{3}P},$ PT wherein ₁P, PT ₂P, PT ₃P representes three shift lengths of camera respectively, after calculating, obtains [λ ₁₀/ PT ₁P (λ ₂₀-λ ₁₀)/PT ₂P (λ ₃₀-λ ₂₀-λ ₁₀)/PT ₃P] [X Y Z] ^T=1, through understanding three translation distances of camera, just can derive three scale factor λ ₁₀, λ ₂₀, λ ₃₀, recover planar structure information then, its computation process is linear.

Calculate the intersection point on three planes at last, can recover the three-dimensional structure image.

Description of drawings

Figure is 3-D view flow process figure of the present invention.

Embodiment

For clear statement implementation process of the present invention, specifically describe the rejuvenation of 3-D view below.

As shown in the figurely come the analogue camera imaging process with the pin-hole imaging model, object point P is imaged on picture point P ' through photocentre, and d is a focal length.Utilize three translation distances of active vision platform measuring and record camera then, as the P of camera from beginning ₀, take turns doing the quadrature translation motion to P ₁, P ₂, P ₃, utilize position P ₀As world coordinate system, note P ₀To P ₁, be T ₁, P ₁, to P ₂Be T ₂, P ₂, to P ₃Be T ₃, and satisfy

(i=1,2,3, j ≠ i), note P ₁, P ₂, P ₃The position with respect to P ₀Displacement be respectively T ₁₀=T ₁, T ₂₀=T ₁+ T ₂, T ₃₀=T ₁+ T ₂+ T ₃

When carrying out pure flat shifting movement, rotation matrix and infinite distance matrix that view is corresponding are unit matrix, and then

Wherein I representes 3 * 3 unit matrix, utilizes match point can confirm homography matrix H, further draws matrix D after finding the solution scale factor β through the order of matrix D is unique, and then derives N ^TT=d λ.

The recovering step of planar structure: N ^TT ₁₀=d λ ₁₀, N ^TT ₂₀=d λ ₂₀, N ^TT ₃₀=d λ ₃₀, translation vector is carried out normalization can get three set of equations as follows: $\frac{N^T{T}_1}{{\mathrm{PT}}_{1}P}=\frac{d{\mathrm{\λ}}_{10}}{P{T}_{1}P},$ $\frac{N^T{T}_2}{{\mathrm{PT}}_{2}P}=\frac{d({\mathrm{\λ}}_{20}-{\mathrm{\λ}}_{10})}{{\mathrm{PT}}_{2}P},$ $\frac{N^T{T}_3}{{\mathrm{PT}}_{3}P}=\frac{d({\mathrm{\λ}}_{30}-{\mathrm{\λ}}_{20}-{\mathrm{\λ}}_{10})}{{\mathrm{PT}}_{3}P},$ PT wherein ₁P, PT ₂P, PT ₃P representes three shift lengths of camera respectively, considers a new coordinate system, with three quadrature unit vector T ₁/ PT ₁P, T ₂/ PT ₂P, T ₃/ PT ₃P answers and and P as the basic coordinates axle ₀The initial point of coordinate system identical, guarantee new normal vector [the d λ of plane ∏ correspondence under new coordinate system ₁₀/ PT ₁Pd (λ ₂₀-λ ₁₀)/PT ₂Pd (λ ₃₀-λ ₂₀-λ ₁₀)/PT ₃P] ^TBe unit vector, the unit normal vector of plane ∏ under new coordinate system just, thus obtain [d λ ₁₀/ PT ₁Pd (λ ₂₀-λ ₁₀)/PT ₂Pd (λ ₃₀-λ ₂₀-λ ₁₀)/PT ₃P] ^T=d obtains [λ after calculating ₁₀/ PT ₁P (λ ₂₀-λ ₁₀)/PT ₂P (λ ₃₀-λ ₂₀-λ ₁₀)/PT ₃P] [X Y Z] ^T=1, through understanding three translation distances of camera, just can derive three scale factor λ ₁₀, λ ₂₀, λ ₃₀, recover planar structure information then, its computation process is linear.

Calculate the intersection point on three planes at last, can recover 3-D view.

The foregoing description does not limit the present invention in any way, and every employing is equal to the technical scheme that replacement or the mode of equivalent transformation obtain and all drops in protection scope of the present invention.

Claims (5)

1. the image processing method of a three-dimensional camera; It is characterized in that: algorithm only need utilize active vision platform control camera to do three quadrature translation motions, just can recover planar structure information and carry out the Euclidean reconstruction through the distance of image corresponding point and translation motion then.

2. image processing method according to claim 1 is characterized in that: active vision platform control camera is done three quadrature translation motions and is only carried out one group.

3. according to claim 1 and 2 described image processing methods, it is characterized in that: camera is done pure flat shifting movement, and rotation matrix and infinity homographies that two views are corresponding are unit matrix.

4. image processing method according to claim 1, recover planar structure information and be characterised in that: measure three translation distances, utilize the image corresponding point to calculate then and can recover planar structure information, its computation process is linear.

5. image processing method according to claim 1, it is characterized in that of European reconstruction: utilize three quadrature translation distances of active vision platform measuring and record.

Images