CN101998136A - Homography matrix acquisition method as well as image pickup equipment calibrating method and device - Google Patents

Abstract

The invention relates to a homography matrix acquisition method as well as an image pickup equipment calibrating method and device. The image pickup equipment calibrating method comprises the following steps: acquiring at least three images from one scene, respectively extracting characteristic points from each image, and carrying out characteristic point matching on the extracted characteristic points to obtain pixel coordinates of the matched characteristic points corresponding to the same point in the three-dimensional scene space in each image, wherein each image has a relative rotating angle among shooting positions of the image pickup equipment in the shooting process; calculating the fundamental matrix F between any two images and the extremity coordinates e of each image based on the pixel coordinates, and calculating the projection matrix P of the image pickup equipment based on F and e; solving the homography matrix H of which the principal diagonal elements are set to be approximately equal based on the projection matrix P; and measuring and calibrating based on the solution to obtain internal parameters of the image pickup equipment. By using the technical scheme, convenient, simple, fast and precise on-line calibration on parameters of the image pickup equipment can be obtained without depending on reference objects for calibration.

Description

The acquisition methods of homography matrix, the scaling method of picture pick-up device and device

Technical field

The present invention relates to technical field of image processing, be specifically related to the acquisition methods of homography matrix and the scaling method and the device of picture pick-up device.

Background technology

Picture pick-up device is demarcated the geometrical model parameter that picture pick-up device promptly is set.Picture pick-up device geometrical model parameter is used for determining three-dimensional geometry position and its correlation between the image corresponding points of certain point of space object surface.Picture pick-up device geometrical model parameter is also referred to as the picture pick-up device parameter.The picture pick-up device parameter comprises inner parameter and external parameter.

At present picture pick-up device inner parameter scaling method mainly comprises: traditional standardization and from standardization.

The tradition standardization is under certain picture pick-up device geometrical model, carries out image processing at shape, the known calibrated reference of size, to obtain the inner parameter and the external parameter of video camera.The tradition standardization can obtain higher precision, but owing to need to use calibrated reference in the picture pick-up device calibration process, therefore, traditional standardization is not suitable for the application scenarios of complexity such as picture pick-up device need take regular exercise.

From standardization can scene the unknown, and the situation of picture pick-up device arbitrary motion under determine the inner parameter of picture pick-up device.On standardization has utilized angle in projective geometry, there is the quadratic nonlinearity bounding theory between per two width of cloth images, as there being two Kruppa equations etc. between per two width of cloth images, by being retrained, quadratic nonlinearities such as Kruppa equation group directly find the solution, obtain the inner parameter of picture pick-up device.Though from standardization as based on the Kruppa equation can be implemented in the real-time calibration of line from standardization, exist because equation solution difficulty etc. are former thereby stated accuracy that cause is not high and demarcate problem such as implementation procedure complexity.

In addition, all need to use homography matrix H in multiple application such as picture pick-up device demarcation, image splicing, H is generally one 3 * 3 matrix, and the degree of freedom is 8, and H can represent two transformation relations between the projection plane:

$x^{\′}=\mathrm{Hx}=\left[\begin{array}{ccc}{h}_{11}& h_{12}& h_{13}\\ h_{21}& h_{22}& h_{23}\\ h_{31}& h_{32}& h_{33}\end{array}\right]x$

Wherein, x is the homogeneous expression of image coordinate before the conversion, and x ' is the homogeneous expression of image coordinate after the conversion.The acquisition methods of existing homography matrix comprises: manual specify at least 4 pairs of characteristic points to or utilize the mode of taking template image to determine that at least 4 pairs of characteristic points are right, afterwards, utilize characteristic point to setting up equation and finding the solution to obtain homography matrix.Utilize before the known transform and conversion after a characteristic point on the image coordinate is obtained two equations:

$x^{\′}=\frac{h_{11}x+h_{12}y+h_{13}}{h_{31}x+h_{32}y+h_{33}}$ $y^{\′}=\frac{h_{21}x+h_{22}y+h_{23}}{h_{31}x+h_{32}y+h_{33}}$

Because the degree of freedom of H is 8, therefore minimum need be by 4 pairs of characteristic points to setting up 8 equations, and after at least 8 equations are found the solution, just can obtain homography matrix H.

The acquisition methods of above-mentioned homography matrix is owing to need manual specific characteristic point or utilize the shooting template image to find characteristic point, the implementation procedure complexity.

Summary of the invention

Embodiment of the present invention provides the acquisition methods of homography matrix, the scaling method and the device of picture pick-up device, simplified the implementation procedure of obtaining homography matrix, the inner parameter of real-time calibration picture pick-up device that can be online, and, can determine the inner parameter of picture pick-up device simply, fast and accurately.

The acquisition methods of the homography matrix that embodiment of the present invention provides comprises:

Obtain at least three width of cloth images of Same Scene, have relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

Calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image according to the pixel coordinate of the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point, and calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

According to projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution, to obtain homography matrix H _Ai

The scaling method of the picture pick-up device that embodiment of the present invention provides comprises:

Obtain at least three width of cloth images of Same Scene, have relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

Calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image according to the pixel coordinate of the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point, calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

According to projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution;

Measure demarcation according to solving result, to obtain the inner parameter of picture pick-up device.

The homography matrix deriving means that embodiment of the present invention provides comprises:

Image collection module is used to obtain at least three width of cloth images of Same Scene, has relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Extract matching module, described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

The projection matrix module, be used for pixel coordinate according to the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point and calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image, calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

The homography matrix module is used for according to described projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution, the homography matrix H of acquisition is found the solution in output _Ai

The caliberating device of the picture pick-up device that embodiment of the present invention provides comprises:

Image collection module is used to obtain at least three width of cloth images of Same Scene, has relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Extract matching module, be used for described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

The projection matrix module, be used for pixel coordinate according to the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point and calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image, calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

The homography matrix module is according to described projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution;

The tolerance demarcating module is used for measuring demarcation according to the solving result of homography matrix module, to obtain the inner parameter of picture pick-up device.

Description by technique scheme as can be known, the embodiment of the invention has been simplified the implementation procedure of obtaining homography matrix, thereby has simplified the multiple application based on homography matrix; The embodiment of the invention can be under the situation that does not rely on calibrated reference, the inner parameter of easy, quick, accurate real-time on-line proving picture pick-up device.

Description of drawings

Fig. 1 is the scaling method flow chart of the acquisition methods that includes homography matrix of the embodiment of the invention one at interior picture pick-up device;

Fig. 2 is the image sequence schematic diagram one of the embodiment of the invention one;

Fig. 3 is the image sequence schematic diagram two of the embodiment of the invention one;

Fig. 4 is the deriving means structural representation of the homography matrix of the embodiment of the invention two;

Fig. 5 is the caliberating device structural representation of the picture pick-up device of the embodiment of the invention three.

Embodiment

The embodiment of the invention proposes the picture pick-up device calibration technique that a kind of homography matrix obtains technology and rotates based on low-angle.Homography matrix obtain with picture pick-up device calibration technique scheme in, at least three width of cloth images of need to obtain under the Same Scene, picture pick-up device (as video camera) being taken with different shooting angles, for example under the condition that the inner parameter of picture pick-up device remains unchanged, picture pick-up device is taken at least three width of cloth images of different shooting angles under the Same Scene.In shooting process, but the picture pick-up device free shift, still, when taking each width of cloth image, relative rotation angle between the camera site of picture pick-up device should be less, and the relative rotation angle here is at arbitrary axle among X-axis, Y-axis and the Z axle three or any multiaxis.As between any two width of cloth images at three mutual vertical coordinate axle X of picture pick-up device coordinate system, the relative rotation angle on Y and the Z all should be controlled within (15 °, 15 °).Can with the photocentre of picture pick-up device the initial point of X-axis, Y-axis and Z axle.In addition, can think that the clockwise anglec of rotation is a negative angle, the anticlockwise anglec of rotation is a positive-angle.After adopting aforesaid way to obtain at least three width of cloth images, the characteristic point of each width of cloth image is extracted and matching treatment, afterwards, obtain the projection matrix of picture pick-up device by the characteristic point after the matching treatment, and utilize projection matrix that the element of leading diagonal is set to approximately equalised homography matrix and find the solution, the obtain manner of above-mentioned homography matrix can be applied in the multiple application such as picture pick-up device demarcation.

The acquisition methods that includes homography matrix below in conjunction with 1 pair of embodiment of the invention of accompanying drawing describes at interior picture pick-up device scaling method.

Among Fig. 1, step 1, obtain at least three width of cloth images under the Same Scene.

Picture pick-up device is by taking at least three width of cloth images that obtain above-mentioned Same Scene.In the process of taking each width of cloth image, the picture pick-up device inner parameter can remain unchanged, and picture pick-up device can free shift, but, when taking each width of cloth image, the relative rotation angle between the camera site of picture pick-up device should be less, and the camera position when taking any two width of cloth images at picture pick-up device is at three vertical coordinate axle X of picture pick-up device coordinate system, relative rotation angle on Y and the Z should be controlled within (15 °, 15 °).That is to say that the relative rotation angle between any two width of cloth images should be less, the relative rotation angle between any two width of cloth images can be controlled within (15 °, 15 °).In the shooting process of reality, picture pick-up device obtains aforesaid at least three width of cloth images and is easy to realize.

A specific implementation process of step 1 is: remain unchanged in picture pick-up device inner parameter matrix K, picture pick-up device can translation and can make between any two width of cloth images relative rotation angle at (15 °, 15 °) in situation under, take at least three width of cloth images, the image that the image sequence that setting photographs comprises is: I ₁, I ₂... I _M, M wherein 〉=3.A concrete image sequence as shown in Figure 2, four width of cloth images that Fig. 2 photographs at certain scene for picture pick-up device satisfy above-mentioned low-angle rotating condition between any two width of cloth images in this four width of cloth image.Another concrete image sequence as shown in Figure 3, Fig. 3 four width of cloth images that to be picture pick-up device take at another scene satisfy the low-angle rotating condition between any two images in this four width of cloth image equally.Image illustrated in figures 1 and 2 is PAL SD (720 * 576) image.

An object lesson of above-mentioned picture pick-up device inner parameter matrix K is:

$K=\left[\begin{array}{ccc}{k}_u& s& p_u\\ 0& k_v& p_v\\ 0& 0& 1\end{array}\right],$ Formula (1)

K in the formula (1) is the upper triangular matrix of picture pick-up device inner parameter, wherein, and k _uBe that image is the multiplication factor of unit with the pixel at u direction (laterally), k _vBe that image is the multiplication factor of unit with the pixel in v direction (vertically), s is the distortion in images factor, p _u, p _vBe to be the coordinate of the figure principal point of unit with the pixel.

Above-mentioned parameter k _uAnd k _vWith the focal length of picture pick-up device close getting in touch arranged, for example, in the photosensitive array of picture pick-up device, comprise (k under the situation of square pixel _u=k _v), if s=0, then k _uAnd k _vPromptly be to be the focal length of the picture pick-up device of unit with the pixel; Again for example, in the photosensitive array of picture pick-up device, comprise under the situation of non-square pixels (such as ccd video camera) k _iBe the ratio of the size of focal distance f and u direction pixel, k _vBe the ratio of the size of focal distance f and v direction pixel.

Determine above-mentioned picture pick-up device inner parameter matrix K and promptly determine the picture pick-up device inner parameter.

Step 2, at every width of cloth image, carry out feature point extraction respectively and handle, by the characteristic point of extracting being mated the pixel coordinate that obtains the same point of corresponding three-dimensional scene space in every width of cloth image.

A specific implementation process of step 2 is: utilize existing feature point extraction technology all to carry out feature point extraction at every width of cloth image and handle, as based on Feature Points Extraction or SIFT (Scale-Invariant Feature Transform, the eigentransformation of yardstick consistency) feature extracting method etc. every width of cloth image carried out feature point extraction respectively handle.Then, obtain the pixel coordinate of corresponding three-dimensional scene space same point in every width of cloth image respectively by Feature Points Matching.Wherein, every width of cloth image characteristic point number of carrying out matching treatment should be at least 8.The characteristic point number of mating in the image is many more, and the constraint that then provides is many more, thereby can make that the inherence constraint corresponding relation between the image is accurate more, finally makes the calibration result of picture pick-up device accurate more.

Step 3, the projection matrix P that utilizes the pixel coordinate of the same point of corresponding three-dimensional scene space in every width of cloth image to obtain picture pick-up device (are P _Pi, wherein i=1 2......M), promptly utilizes pixel coordinate to carry out the Projection surveying of picture pick-up device.

Utilizing pixel coordinate to obtain in the process of projection matrix P, can be earlier the pixel coordinate of the characteristic point after the above-mentioned matching treatment be carried out normalized transformation, the pixel coordinate of the characteristic point after promptly handling at each width of cloth images match is unified conversion, generate normalized pixel coordinate, be set to approximately equalised approximate condition so that satisfy the element of the leading diagonal of projection matrix P.That is to say, normalized pixel coordinate is carried out Projection surveying, to obtain the picture pick-up device projection matrix of every width of cloth image in the backprojection reconstruction space.A specific implementation process that generates normalized pixel coordinate in the step 3 is: for the approximately equalised condition of element of the leading diagonal that makes projection matrix P is set up, need carry out preliminary treatment to the original pixels coordinate of the characteristic point after the matching treatment in the image, by unified conversion, generate normalized pixel coordinate.A concrete conversion example is:

With each the characteristic point original pixels coordinate u=(u, v, 1) in each image ^TBe transformed to standardization pixel coordinate u '=(u ', v ', 1) by u '=Tu ^T, standardization pixel coordinate u ' is applied in the follow-up layering calibration process, and promptly characteristic point original pixels coordinate u directly is not applied in the follow-up processing procedure.

Above-mentioned $T=\left(\begin{array}{ccc}\frac{1}{2c_u}& 0& -\frac{1}{2}\\ 0& \frac{1}{2c_v}& -\frac{1}{2}\\ 0& 0& 1\end{array}\right)$ Formula (2)

In formula (2), c _uAnd c _vIt is the centre coordinate of image.

Alignd with the photo coordinate system initial point in the center of each image, and two components of new characteristic point coordinate (pixel coordinate promptly standardizes) all drop in (1,1) scope.

According to picture pick-up device perspective projection model, be tied to form upright just like ShiShimonoseki:

U～PX～K[R|-Rt] X formula (3)

In the formula (3), u is the characteristic point coordinate (homogeneous form) on the image, X=(x, y, z, 1) ^TBe the coordinate (homogeneous form) of characteristic point in the three-dimensional scenic space, P is the projection matrix of image correspondence, K as shown in Equation (1), R is respectively the spin matrix and the translation vector of the relative world coordinate system of picture pick-up device coordinate system with t.

The pixel coordinate of the characteristic point of image will be tied to form upright just like ShiShimonoseki through behind the normalized transformation:

U '=Tu～TK[R|-Rt] X=K ' [R|-Rt] X formula (4)

From formula (4) as can be seen, if based on the perspective projection equation, can solve K '.

Because there are following relation in K and K ':

K '=TK formula (5)

Therefore, can obtain in conjunction with formula (1), formula (2) and formula (5):

$K^{\′}=\left(\begin{array}{ccc}{k}_u^{\′}& s^{\′}& p_u^{\′}\\ 0& k_u^{\′}& p_v^{\′}\\ 0& 0& 1\end{array}\right)=\left(\begin{array}{ccc}\frac{k_u}{2c_u}& \frac{s}{2c_u}& \frac{p_u-c_u}{2c_u}\\ 0& \frac{k_v}{2c_v}& \frac{p_v-c_v}{2c_v}\\ 0& 0& 1\end{array}\right)$ Formula (6)

From formula (6) as can be seen, solving on the basis of K ', utilizing formula (6) can obtain original picture pick-up device inner parameter matrix K again based on formula (4).

Obtain the projection matrix P of the picture pick-up device of every width of cloth image in the backprojection reconstruction space _Pi(i=1 2......M) also can be called Projection surveying, at projection matrix P _PiIn, M 〉=3.

In step 3, can come the projection matrix of the picture pick-up device of computed image by basis matrix and limit.

A kind of example of implementation of projection matrix of picture pick-up device of computed image of simplification is: the world coordinate system and the photographic images I that set photographed scene ₁The time the camera coordinates system align, like this, image I ₁Can be considered to reference picture, and the projection matrix of following picture pick-up device is set up:

P _P1=[I _{3 * 3}| 0 ₃] formula (7)

Projection matrix P for picture pick-up device _PiThe method for solving of (i＞1) can be as follows:

At first, calculate reference picture I ₁With image I _iBetween basis matrix F _1iFor example, according to image I ₁And image I _iOn the corresponding relation of standardization pixel coordinate of coupling, use 8 algorithms to obtain basis matrix F between two images _1i

Secondly, to F _1iForce contraction to handle, form new basis matrix F ' _1iFor example, according to basis matrix F _1iCharacter to F _1iDo further processing, make F _1iSatisfy the contraction constraint, i.e. rank (F _1i)=2.A specific implementation process of forcing contraction to handle is:

Calculate F _1iSVD decompose, the result is:

$F_{1i}=U_{F_{1i}}{D}_{F_{1i}}{V}_{F_{1i}}^T,$ Formula (8)

Wherein,

By F _1iSingular value form.

Will

Middle minimum singular value is set to 0, forms a new matrix

Order

$F_{1i}^{\′}=U_{F_{1i}}{D}_{F_{1i}}^{\′}{V}_{F_{1i}}^T;$ Formula (9)

F ' _1iBe the image I of being asked ₁With image I _iBetween new basis matrix.

Once more, computed image I _iLimit (epipole) e _1iCoordinate.The limit e here _1iPromptly at photographic images I ₁The time video camera photocentre in image I _iOn projection.According to the relevant knowledge of utmost point geometry, have following homogeneous equation system to set up:

$F_{1i}^{\′T}{e}_{1i}=0;$ Formula (10)

Because F ' _1iBe contraction, therefore, separating of formula (10) is e _1iValue be unique, and e _1iWith Last part of decomposing of SVD

Last row proportional.

At last, calculate the projection matrix P of picture pick-up device _Pi(i＞1).Determine the projection matrix P of picture pick-up device according to the relevant knowledge of selected world coordinate system of photographed scene and projective transformation _Pi(i＞1) is:

P _Pi=[H _Pi| e _1i] (i＞1) formula (11)

In formula (11):

H _Pi=[e _1i] _*F ' _1i+ e _1iπ ^TFormula (12)

In the formula (12), π is non-arbitrarily 0 column vector, as π=[1,1,1] ^T[e _1i] _*Be by e _1iThe antisymmetric matrix that generates makes e _1i=(e ₁, e ₂, e ₃) ^T, then have:

${\left[e_{1i}\right]}_{\&times;}=\left[\begin{array}{ccc}0& -e_3& e_2\\ {\mathrm{<figure-callout\; id="3"\; label="e"\; filenames="H2009100909988E0000011.png"\; state="\{\{state\}\}">e</figure-callout>}}_3& 0& -e_1\\ -e_2& {\mathrm{<figure-callout\; id="1"\; label="e"\; filenames="H2009100909988E0000011.png"\; state="\{\{state\}\}">e</figure-callout>}}_1& 0\end{array}\right],$ Formula (13)

Obtained the projection matrix P of the picture pick-up device of every width of cloth image in the backprojection reconstruction space by foregoing description _Pi(i=1,2......M), wherein Projection surveying has promptly been realized in M 〉=3.

Step 4, according to projection matrix P the element of leading diagonal to be set to approximately equalised homography matrix H (be H _Ai, wherein i=1 2......M) finds the solution, and promptly the projection matrix to the picture pick-up device of above-mentioned acquisition carries out affine demarcation.

According to the characteristics of low-angle rotation between the image, can derive the approximately equal relation between the leading diagonal element of the homography matrix relevant, and determine the parameter of infinity reference planes in the backprojection reconstruction space thus with affine demarcation.That is to say that the target of affine demarcation is that (this parameter is with image I for the definite parameter of infinity reference planes in the backprojection reconstruction space ₁The position as the reference position), promptly determine π _∞=(π ₁, π ₂, π ₃) ^T, and determine the homography matrix H relevant thus with affine demarcation _Ai

Derivation to the relation of the approximately equal between the leading diagonal element of above-mentioned homography matrix describes below.

Can know following two formula establishment according to projection in the layering demarcation, relation affine and that tolerance is demarcated:

$P_{\mathrm{Ai}}~P_{\mathrm{Pi}}{T}_{\mathrm{PA}}^{-1}~[H_{\mathrm{Pi}}-e_{1i}{\mathrm{\&pi;}}_{\&infin;}^T|e_{1i}]$ Formula (14)

P _Ai～P _MiT _AM～[K ' R _MiK ' ^-1| K ' t _Mi] formula (15)

In formula (14) and formula (15), P _Ai, P _MiBe respectively affine transformation matrix and metric transformation matrix, T _PABe the transition matrix that projection matrix is carried out the transition to affine transformation matrix, T _AMBe the transition matrix that affine transformation matrix is carried out the transition to the metric transformation matrix, R _MiAnd t _MiBe respectively applied for token image I _iWith respect to reference picture I ₁The spin matrix in orientation and translation vector, and,

$T_{\mathrm{PA}}~\left(\begin{array}{cc}{I}_{3\&times;3}& 0_3\\ {\mathrm{\&pi;}}_{\&infin;}^{\mathrm{<figure-callout\; id="1"\; label="T"\; filenames="H2009100909988E0000011.png"\; state="\{\{state\}\}">T</figure-callout>}}& 1\end{array}\right)$ Formula (16)

$T_{\mathrm{AM}}~\left(\begin{array}{cc}{K}^{\&prime;-1}& 0_3\\ 0_{\&infin;}^{\mathrm{<figure-callout\; id="1"\; label="T"\; filenames="H2009100909988E0000011.png"\; state="\{\{state\}\}">T</figure-callout>}}& 1\end{array}\right)$ Formula (17)

$R_{\mathrm{Mi}}~\left(\begin{array}{ccc}{r}_{11}& r_{12}& r_{13}\\ r_{21}& r_{22}& r_{23}\\ r_{31}& r_{32}& r_{33}\end{array}\right)$ Formula (18)

Can obtain following important relationship from the first three columns of above-mentioned formula (14) and (15):

$H_{\mathrm{Pi}}-e_{1i}{\mathrm{\&pi;}}_{\&infin;}^T~K^{\&prime;}{R}_{\mathrm{Mi}}{K}^{\&prime;-1}$ Formula (19)

Can make $H_{\mathrm{Ai}}=H_{\mathrm{Pi}}-e_{1i}{\mathrm{\&pi;}}_{\&infin;}^T,$ Formula (20)

In formula (20), H _AiBe at reference picture I with the point on the plane at infinity ₁On projective transformation in image I _iOn the homography matrix of projection.Formula (19) left side is H _AiComprised unknown quantity π _∞, and the concrete calculating in formula (19) the right can get:

$\mathrm{Diag}\left(H_{\mathrm{Ai}}\right)=\left(\begin{array}{c}{r}_{11}+\frac{s^{\&prime;}}{k_u^{\&prime;}}{r}_{21}+\frac{p_u^{\&prime;}}{k_u^{\&prime;}}{r}_{31}\\ r_{22}+\frac{p_v^{\&prime;}}{k_v^{\&prime;}}{r}_{32}-\frac{s^{\&prime;}}{k_u^{\&prime;}}{r}_{21}-\frac{s^{\&prime;}{p}_v^{\&prime;}}{k_u^{\&prime;}{k}_v^{\&prime;}}{r}_{32}\\ r_{33}+\frac{s^{\&prime;}{p}_v^{\&prime;}}{k_u^{\&prime;}{k}_v^{\&prime;}}{r}_{32}-\frac{p_u^{\&prime;}}{k_u^{\&prime;}}{r}_{31}-\frac{p_v^{\&prime;}}{k_v^{\&prime;}}{r}_{32}\end{array}\right)$ Formula (21)

Diag (H _Ai) expression H _AiThe vector formed of leading diagonal element.

Utilize following two two conclusions analyzing acquisition to carry out change process to formula (21), these two analyses are respectively:

Analyze 1, to R _MiAnalysis.

R _MiCan represent with anglec of rotation α, β, γ, thereby can obtain around x, y, three reference axis of z:

$\mathrm{Diag}\left(R_{\mathrm{Mi}}\right)=\left(\begin{array}{c}{r}_{11}\\ r_{22}\\ r_{33}\end{array}\right)=\left(\begin{array}{c}\cos \mathrm{\&beta;}\cos \mathrm{\&gamma;}\\ \cos \mathrm{\&alpha;}\cos \mathrm{\&gamma;}+\sin \mathrm{\&alpha;}\sin \mathrm{\&beta;}\sin \mathrm{\&gamma;}\\ \cos \mathrm{\&alpha;}\cos \mathrm{\&beta;}\end{array}\right)$ Formula (22)

Work as image I _iWith reference picture I ₁When the relative rotation angle that makes progress three reference axis satisfies the low-angle condition, spin matrix R _MiElement on the leading diagonal and 1 is very approaching, and other element is far smaller than element on the leading diagonal, promptly much smaller than 1.For example, when α ∈ [10 °, 10 °], β ∈ [10 °, 10 °] and γ ∈ [10 °, 10 °], have following formula to set up:

0.9689≤r ₁₁≤ 1 formula (23)

0.9646≤r ₂₂≤ 1 formula (24)

0.9689≤r ₃₃≤ 1 formula (25)

| r _Ij|≤0.2475 (formula (26) of i ≠ j)

Can get conclusion 1 from above-mentioned formula (23) to formula (26): can think R approx _MiElement on the leading diagonal is 1, and consequent worst error is (1-0.9646)/0.9646=3.67%.

Analyze 2, according to the analysis of formula (6) to K '.

From original inner parameter matrix K as can be known, the principal point position (p in the inner parameter that calibrates _u, p _v) and desirable principal point position (c _u, c _v) should be more approaching; k _u, k _vWith p _u, p _vShould be from numerical value in same substantially magnitude; And than k _u, k _v, p _uAnd p _vThese parameters, distortion factor s is a very little number.Can get conclusion 2 thus: for the parameter among the K ', with k ' _u, k ' _vCompare p ' _u, p ' _vAnd s ' is very little.

In conjunction with above-mentioned conclusion 1 and conclusion 2 as can be known: in image I _iWith respect to reference picture I ₁Upwards satisfy under the situation of low-angle rotating condition three reference axis, can be similar to and think homography matrix H _AiThe leading diagonal element all be 1, that is:

Diag (H _Ai)～[1,1,1] ^TFormula (27)

Can derive homography matrix H thus _AiThe leading diagonal element between approximately equal relation.

Hence one can see that, when i ≠ 1, can obtain following two separate formula:

H _Ai(1,1) ≈ H _Ai(2,2) formula (28)

H _Ai(1,1) ≈ H _Ai(3,3) formula (29)

Can construct the following π of comprising according to formula (20) and formula (11) _∞Equation:

$\left[\begin{array}{ccc}-P_{\mathrm{Pi}}\left(\mathrm{1,4}\right)& P_{\mathrm{Pi}}\left(\mathrm{2,4}\right)& 0\\ -P_{\mathrm{Pi}}\left(\mathrm{1,4}\right)& 0& P_{\mathrm{Pi}}\left(\mathrm{3,4}\right)\end{array}\right]{\mathrm{\&pi;}}_{\&infin;}=\left[\begin{array}{c}{P}_{\mathrm{Pi}}\left(\mathrm{2,2}\right)-P_{\mathrm{Pi}}\left(\mathrm{1,1}\right)\\ P_{\mathrm{Pi}}\left(\mathrm{3,3}\right)-P_{\mathrm{Pi}}\left(\mathrm{1,1}\right)\end{array}\right]$ Formula (30)

P _Pi(m, n) representing matrix P _PiIn the element of the capable n of m row.

Can obtain from foregoing description, homography matrix H being found the solution according to projection matrix P of step 4 is that the example of a concrete processing procedure of affine demarcation is:

At first, obtain the position π of plane at infinity in the backprojection reconstruction space _∞Because two and plural H _Ai(i.e. the above image of three width of cloth images and three width of cloth) can construct the overdetermined equation group on the basis of formula (30), therefore, use least square method can solve the position π of plane at infinity in the backprojection reconstruction space _∞

Secondly, find the solution homography matrix H _AiUtilize the π of above-mentioned acquisition _∞, and formula (20) can solve each homography matrix H _Ai(i＞1).

At last, standardization homography matrix H _AiSo far, successfully obtained homography matrix.The homography matrix that obtains goes in the multiple application.

At each H _Ai(i＞1) carries out H ' _Ai=α _iH _AiHandle, make det (H ' _Ai)=1.Hence one can see that,

Thereby H _AiTransform to and have the unit determinant:

$H_{\mathrm{Ai}}^{\&prime;}=\sqrt[3]{1/\det \left(H_{\mathrm{Ai}}\right)}\&CenterDot;H_{\mathrm{Ai}}.$ Formula (31)

And formula (19) is changed to:

H′ _Ai＝K′R _MiK′ ^-1。Formula (32)

Step 5, the parameter in the backprojection reconstruction space is measured demarcation, promptly determine picture pick-up device inner parameter matrix K according to homography matrix.

The object lesson that tolerance is demarcated is:

At first, find the solution the picture pick-up device inner parameter matrix K relevant with the pixel coordinate that standardizes '.

Can get according to formula (32): K ' ^-1H ' _AiK '=R _MiCharacter according to spin matrix

Can get:

Can obtain thus:

$K^{\&prime;}{K}^{\&prime;T}=H_{\mathrm{Ai}}^{\&prime;}\left(K^{\&prime;}{K}^{\&prime;T}\right)H_{\mathrm{Ai}}^{\&prime;T}.$ Formula (33)

Order $C=K^{\&prime;}{K}^{\&prime;T}=\left(\begin{array}{ccc}a& b& c\\ b& d& e\\ c& e& f\end{array}\right),$ Formula (34)

In conjunction with formula (33) and formula (34) as can be known: formula (33) is from H ' _AiThe system of linear equations that contains six separate unknown number a, b, c, d, e and f among the C that generates.After eliminating redundant equation, formula (33) is become by 6 systems of homogeneous linear equations.

By M-1 H ' _Ai(M＞=3) can form following overdetermination homogeneous linear equations system:

XC '=0; Formula (35)

In formula (35), C '=(a, b, c, d, e, f) ^TThe vector of being made up of the independent entry of C, and X is the matrix of one 6 (M-1) * 6 uses SVD to decompose or jacobi method can be found the solution this equation system.

After solving C ', according to formula (34), utilize the Cholesky decomposition method can obtain K '.

Secondly, after solving K ', solve original picture pick-up device inner parameter matrix K by formula (5).

Can get by formula (5):

K=T ^-1K ' formula (36)

Under the situation of known T and K ', utilize formula (36) can obtain original picture pick-up device inner parameter matrix K.

The scaling method that adopts the embodiment of the invention to describe, as shown in table 1 at the result that Fig. 2 and Fig. 3 demarcate.

Table 1

The scaling method of the embodiment of the invention can be called the picture pick-up device layering self-calibrating method based on the low-angle rotation, low-angle rotation is that the anglec of rotation relative between the image is less, and layering is from demarcating promptly according to projection, affine and measure three and demarcate picture pick-up device that levels carry out from demarcating.This scaling method goes for the camera calibration in computer vision and the close-range photography measurement.The scaling method of the embodiment of the invention can be in the process of computer vision and photogrammetric task, under the situation that does not rely on calibrated reference, and easy, on-line proving picture pick-up device parameter fast and accurately.

Through the above description of the embodiments, those skilled in the art can be well understood to the present invention and can realize by the mode that software adds essential hardware platform, can certainly all implement, but the former is better execution mode under a lot of situation by hardware.Based on such understanding, all or part of can the embodying that technical scheme of the present invention contributes to background technology with the form of software product, this computer software product can be stored in the storage medium, as ROM/RAM, magnetic disc, CD etc., comprise that some instructions are with so that a computer equipment (can be a personal computer, server, the perhaps network equipment etc.) carry out the described method of some part of each embodiment of the present invention or embodiment.

The deriving means of embodiment two, homography matrix.

The structure of this device as shown in Figure 4.The deriving means of the homography matrix among Fig. 4 comprises: image collection module 400, extraction matching module 410, projection matrix module 420 and homography matrix module 430.

Image collection module 400 is used to obtain at least three width of cloth images of Same Scene, has relative rotation angle between the camera site of each width of cloth image picture pick-up device when taking.Image collection module 400 can be obtained image by the mode of oneself taking, and at this moment, image collection module 400 self promptly is the photographing section of picture pick-up device.Image collection module 400 can not obtained image by the mode of oneself taking yet, and for example, image collection module 400 can be graded from the shoot part of picture pick-up device and be obtained image, and at this moment, image collection module 400 self can not possess image camera function.

Each image that image collection module 400 is obtained is in shooting process, the inner parameter of picture pick-up device can remain unchanged, and picture pick-up device can free shift, but when taking each width of cloth image, the relative rotation angle between the camera site of picture pick-up device should be less, camera position when taking any two width of cloth images at picture pick-up device is at three vertical coordinate axle X of picture pick-up device coordinate system, Y, the relative rotation angle on the Z should be controlled within (15 °, 15 °).That is to say that the relative rotation angle between any two width of cloth images should be less, the relative rotation angle between any two width of cloth images can be controlled within (15 °, 15 °).In the shooting process of reality, picture pick-up device obtains aforesaid at least three width of cloth images and is easy to realize.Object lesson of the image sequence of the inner parameter matrix K of picture pick-up device and shooting or the like is as the description of above-mentioned method embodiment.In this no longer repeat specification.

Extract matching module 410, each width of cloth image that is used for image collection module 400 is got access to carries out feature point extraction respectively, and the characteristic point of extracting is carried out Feature Points Matching, to obtain the pixel coordinate of corresponding three-dimensional scene space same point in each width of cloth image.

Extracting matching module 410 can utilize existing feature point extraction technology all to carry out the feature point extraction processing at every width of cloth image, as based on Feature Points Extraction or SIFT (Scale-invariant feature transform, the eigentransformation of yardstick consistency) feature extracting method etc. extracts 410 pairs of every width of cloth images of matching module and carries out feature point extraction respectively and handle.Then, extract matching module 410 obtains corresponding three-dimensional scene space same point in every width of cloth image respectively by Feature Points Matching pixel coordinate.Wherein, extract the characteristic point number that 410 pairs of every width of cloth images of matching module carry out matching treatment and should be at least 8.The characteristic point number of mating in the image is many more, and the constraint that then provides is many more, thereby can make that the inherence constraint corresponding relation between the image is accurate more, finally makes the calibration result of picture pick-up device accurate more.

Projection matrix module 420 is used for calculating the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image according to extracting pixel coordinate that matching module 410 obtains, and utilizes F and e to calculate the projection matrix P of picture pick-up device.Be that projection matrix module 420 utilizes pixel coordinate to carry out the Projection surveying of picture pick-up device.

Utilizing pixel coordinate to obtain in the process of projection matrix P, projection matrix module 420 can be earlier carried out normalized transformation with the pixel coordinate of the characteristic point after the above-mentioned matching treatment, the pixel coordinate of the characteristic point after promptly handling at each width of cloth images match is unified conversion, generate normalized pixel coordinate, be set to approximately equalised approximate condition so that satisfy the element of the leading diagonal of projection matrix P.That is to say that 420 pairs of normalized pixel coordinates of projection matrix module carry out Projection surveying, to obtain the picture pick-up device projection matrix of every width of cloth image in the backprojection reconstruction space.Projection matrix module 420 can be carried out preliminary treatment to the original pixels coordinate of the characteristic point after the matching treatment in the image, by unified conversion, generates normalized pixel coordinate.Projection matrix module 420 carry out conversion with implementation of the projection matrix of the picture pick-up device of the object lesson that generates normalized pixel coordinate and projection matrix module 420 computed image or the like as the description among the above-mentioned method embodiment.

The mode that homography matrix H finds the solution has multiple, if homography matrix module 430, be used for the projection matrix P that calculates according to projection matrix module 420 and the element of leading diagonal is set to approximately equalised homography matrix H finds the solution, and the homography matrix of acquisition is found the solution in output.The operation that homography matrix module 430 is carried out also can be called affine demarcation.No longer the derivation of the relation of the approximately equal between the leading diagonal element of homography matrix is carried out repeat specification in the present embodiment.

Cell matrix module 430 utilizes projection matrix P that the element of leading diagonal is set to the approximately equalised π of passing through _∞Realize finding the solution then of homography matrix H, homography matrix module 430 can comprise: first submodule 431 and second submodule 432.That is to say that the target of the affine demarcation of first submodule 431 and second submodule 432 is that (this parameter is with image I for the definite parameter of infinity reference planes in the backprojection reconstruction space ₁The position as the reference position), promptly determine π _∞=(π ₁, π ₂, π ₃) ^T, and determine the homography matrix H relevant thus with affine demarcation _Ai

First submodule 431, the projection matrix P that is used for calculating according to projection matrix module 420 calculates plane at infinity at the position in backprojection reconstruction space π _∞Because two and plural H _Ai(i.e. the above image of three width of cloth images and three width of cloth) can construct the overdetermined equation group on the basis of formula (30), therefore, first submodule 431 can use least square method can solve the position π of plane at infinity in the backprojection reconstruction space _∞

Second submodule 432 is used to the π that utilizes first submodule 431 to calculate _∞The element of finding the solution the leading diagonal between two width of cloth images is set to approximately equalised homography matrix H.The π that second submodule 432 can utilize first submodule 431 to obtain _∞, and formula (20) can solve each homography matrix H _Ai(i＞1).The second submodule 432 homography matrix H that can also standardize _AiSecond submodule 432 can be exported standardization homography matrix H _AiThe homography matrix of second submodule, 432 outputs goes in the multiple application.

The caliberating device of embodiment three, picture pick-up device.

This device can be picture pick-up device, also can be the part unit that is provided with in the picture pick-up device.An object lesson of this apparatus structure as shown in Figure 5.

The caliberating device of the picture pick-up device among Fig. 5 comprises: image collection module 500, extraction matching module 510, projection matrix module 520, homography matrix module 530 and tolerance demarcating module 540.

Image collection module 500 is used to obtain at least three width of cloth images of Same Scene, has relative rotation angle between the camera site of each width of cloth image picture pick-up device when taking.Image collection module 500 can be obtained image by the mode of oneself taking, and at this moment, image collection module 500 self promptly is the photographing section of picture pick-up device.Image collection module 500 can not obtained image by the mode of oneself taking yet, and for example, image collection module 500 can be graded from the shoot part of picture pick-up device and be obtained image, and at this moment, image collection module 500 self can not possess image camera function.

Each image that image collection module 500 is obtained is in shooting process, the inner parameter of picture pick-up device can remain unchanged, and picture pick-up device can free shift, but when taking each width of cloth image, the relative rotation angle between the camera site of picture pick-up device should be less, camera position when taking any two width of cloth images at picture pick-up device is at three vertical coordinate axle X of picture pick-up device coordinate system, Y, the relative rotation angle on the Z should be controlled within (15 °, 15 °).That is to say that the relative rotation angle between any two width of cloth images should be less, the relative rotation angle between any two width of cloth images can be controlled within (15 °, 15 °).In the shooting process of reality, picture pick-up device obtains aforesaid at least three width of cloth images and is easy to realize.Object lesson of the image sequence of the inner parameter matrix K of picture pick-up device and shooting or the like is as the description of above-mentioned method embodiment.In this no longer repeat specification.

Extract matching module 510, each width of cloth image that is used for image collection module 500 is got access to carries out feature point extraction respectively, and the characteristic point of extracting is carried out Feature Points Matching, to obtain the pixel coordinate of corresponding three-dimensional scene space same point in each width of cloth image.

Extracting matching module 510 can utilize existing feature point extraction technology all to carry out the feature point extraction processing at every width of cloth image, as based on Feature Points Extraction or SIFT (Scale-invariant feature transform, the eigentransformation of yardstick consistency) feature extracting method etc. extracts 510 pairs of every width of cloth images of matching module and carries out feature point extraction respectively and handle.Then, extract matching module 510 obtains corresponding three-dimensional scene space same point in every width of cloth image respectively by Feature Points Matching pixel coordinate.Wherein, extract the characteristic point number that 510 pairs of every width of cloth images of matching module carry out matching treatment and should be at least 8.The characteristic point number of mating in the image is many more, and the constraint that then provides is many more, thereby can make that the inherence constraint corresponding relation between the image is accurate more, finally makes the calibration result of picture pick-up device accurate more.

Projection matrix module 520 is used for calculating the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image according to extracting pixel coordinate that matching module 510 obtains, and utilizes F and e to calculate the projection matrix P of picture pick-up device.Be that projection matrix module 520 utilizes pixel coordinate to carry out the Projection surveying of picture pick-up device.

Utilizing pixel coordinate to obtain in the process of projection matrix P, projection matrix module 520 can be earlier carried out normalized transformation with the pixel coordinate of the characteristic point after the above-mentioned matching treatment, the pixel coordinate of the characteristic point after promptly handling at each width of cloth images match is unified conversion, generate normalized pixel coordinate, be set to approximately equalised approximate condition so that satisfy the element of the leading diagonal of projection matrix P.That is to say that 520 pairs of normalized pixel coordinates of projection matrix module carry out Projection surveying, to obtain the picture pick-up device projection matrix of every width of cloth image in the backprojection reconstruction space.Projection matrix module 520 can be carried out preliminary treatment to the original pixels coordinate of the characteristic point after the matching treatment in the image, by unified conversion, generates normalized pixel coordinate.Projection matrix module 520 carry out conversion with implementation of the projection matrix of the picture pick-up device of the object lesson that generates normalized pixel coordinate and projection matrix module 520 computed image or the like as the description among the above-mentioned method embodiment.

Homography matrix module 430 is used for the projection matrix P that calculates according to projection matrix module 520 and the element of leading diagonal is set to approximately equalised homography matrix H finds the solution, and the homography matrix of acquisition is found the solution in output.The operation that homography matrix module 430 is carried out also can be called affine demarcation.No longer the derivation of the relation of the approximately equal between the leading diagonal element of homography matrix is carried out repeat specification in the present embodiment.

Cell matrix module 530 utilize projection matrix P to the element of leading diagonal be set to mode that approximately equalised homography matrix H finds the solution have multiple, if pass through π _∞Realize finding the solution then of homography matrix H, homography matrix module 530 can comprise: first submodule 531 and second submodule 532.That is to say that the target of the affine demarcation of first submodule 531 and second submodule 532 is that (this parameter is with image I for the definite parameter of infinity reference planes in the backprojection reconstruction space ₁The position as the reference position), promptly determine π _∞=(π ₁, π ₂, π ₃) ^T, and determine the homography matrix H relevant thus with affine demarcation _Ai

First submodule 531, the projection matrix P that is used for calculating according to projection matrix module 420 calculates plane at infinity at the position in backprojection reconstruction space π _∞Because two and plural H _Ai(i.e. the above image of three width of cloth images and three width of cloth) can construct the overdetermined equation group on the basis of formula (30), therefore, first submodule 531 can use least square method can solve the position π of plane at infinity in the backprojection reconstruction space _∞

Second submodule 532 is used to the π that utilizes first submodule 531 to calculate _∞The element of finding the solution the leading diagonal between two width of cloth images is set to approximately equalised homography matrix H.The π that second submodule 532 can utilize first submodule 531 to obtain _∞, and formula (20) can solve each homography matrix H _Ai(i＞1).The second submodule 532 homography matrix H that can also standardize _AiSecond submodule 532 can be exported standardization homography matrix H _AiThe homography matrix of second submodule, 532 outputs goes in the multiple application.

Tolerance demarcating module 540 is used for measuring demarcation according to the solving result of homography matrix module 530, to obtain the inner parameter of picture pick-up device.

The object lesson that tolerance demarcating module 540 is measured demarcation is: tolerance demarcating module 540 find the solution the picture pick-up device inner parameter matrix K relevant with the pixel coordinate that standardizes ', then, utilize T and K ' according to formula K=T ^-1K ' solves original picture pick-up device inner parameter matrix K.Wherein, tolerance demarcating module 540 obtains the detailed process of K ' such as the description among the above-mentioned method embodiment, in this no longer repeat specification.

Though described the present invention by embodiment, those of ordinary skills know, the present invention has many distortion and variation and do not break away from spirit of the present invention, and the claim of application documents of the present invention comprises these distortion and variation.

Claims (16)

1. the acquisition methods of a homography matrix is characterized in that, comprising:

Obtain at least three width of cloth images of Same Scene, have relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

Calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image according to the pixel coordinate of the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point, and calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

According to projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution, to obtain homography matrix H _Ai

2. the method for claim 1 is characterized in that, the scope in the relative rotation angle of X-axis and/or Y-axis and/or Z axle between the described camera site is-15 ° to 15 °.

3. the method for claim 1 is characterized in that, described at least three width of cloth images that obtain Same Scene comprise:

Under the condition that the inner parameter of picture pick-up device remains unchanged, picture pick-up device is taken at least three width of cloth images of Same Scene.

4. the method for claim 1 is characterized in that, at piece image, described characteristic point of carrying out Feature Points Matching is at least 8.

5. as the described method of arbitrary claim in the claim 1 to 4, it is characterized in that, described according to projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution and comprise:

According to described projection matrix P _PiCalculate the position π of plane at infinity in the backprojection reconstruction space _∞

Utilize described π _∞The element of finding the solution the leading diagonal between two width of cloth images is set to approximately equalised homography matrix H _Ai

6. the scaling method of a picture pick-up device is characterized in that, comprising:

Obtain at least three width of cloth images of Same Scene, have relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

Calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image according to the pixel coordinate of the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point, calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

According to projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution;

Measure demarcation according to solving result, to obtain the inner parameter of picture pick-up device.

7. method as claimed in claim 6 is characterized in that, the scope in the relative rotation angle of X-axis and/or Y-axis and/or Z axle between the described camera site is-15 ° to 15 °.

8. method as claimed in claim 6 is characterized in that, described at least three width of cloth images that obtain Same Scene comprise:

Under the condition that the inner parameter of picture pick-up device remains unchanged, picture pick-up device is taken at least three width of cloth images of Same Scene.

9. method as claimed in claim 6 is characterized in that, at piece image, described characteristic point of carrying out Feature Points Matching is at least 8.

10. method as claimed in claim 6 is characterized in that, and is described according to projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution and comprise:

According to described projection matrix P _PiCalculate the position π of plane at infinity in the backprojection reconstruction space _∞

Utilize described π _∞The element of finding the solution the leading diagonal between two width of cloth images is set to approximately equalised homography matrix H _Ai

11., it is characterized in that described tolerance is demarcated and comprised as the described method of arbitrary claim in the claim 6 to 10:

Utilize at least 2 described homography matrixs to form overdetermination homogeneous linear equations system, find the solution the inner parameter matrix K of the equation group acquisition picture pick-up device in the described equation system.

12. a homography matrix deriving means is characterized in that, comprising:

Image collection module is used to obtain at least three width of cloth images of Same Scene, has relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Extract matching module, described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

The projection matrix module, be used for pixel coordinate according to the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point and calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image, calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

The homography matrix module is used for according to described projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution, the homography matrix H of acquisition is found the solution in output _Ai

13. device as claimed in claim 12 is characterized in that, the homography matrix module comprises:

First submodule is used for according to described projection matrix P _PiCalculate the position π of plane at infinity in the backprojection reconstruction space _∞

Second submodule is used to utilize described π _∞The element of finding the solution the leading diagonal between two width of cloth images is set to approximately equalised homography matrix H _Ai

14. the caliberating device of a picture pick-up device is characterized in that, comprising:

Image collection module is used to obtain at least three width of cloth images of Same Scene, has relative rotation angle between the camera site of described each width of cloth image picture pick-up device when taking;

Extract matching module, be used for described each width of cloth image is carried out feature point extraction respectively, and the characteristic point of described extraction is carried out Feature Points Matching, to obtain the pixel coordinate of coupling characteristic point corresponding three-dimensional scene space same point in each width of cloth image afterwards;

The projection matrix module, be used for pixel coordinate according to the characteristic point of each width of cloth image of described corresponding three-dimensional scene space same point and calculate the basis matrix F between any two width of cloth images and the limit coordinate e of every width of cloth image, calculate the projection matrix P of picture pick-up device by described F and e _Pi, i=1 wherein, 2......M, M are the described amount of images of obtaining;

The homography matrix module is according to described projection matrix P _PiElement to leading diagonal is set to approximately equalised homography matrix H _AiFind the solution;

The tolerance demarcating module is used for measuring demarcation according to the solving result of homography matrix module, to obtain the inner parameter of picture pick-up device.

15. device as claimed in claim 14 is characterized in that, the homography matrix module comprises:

First submodule is used for according to described projection matrix H _AiCalculate the position π of plane at infinity in the backprojection reconstruction space _∞

Second submodule is used to utilize described π _∞The element of finding the solution the leading diagonal between two width of cloth images is set to approximately equalised homography matrix H _Ai

16. as claim 14 or 15 described devices, it is characterized in that, at least 2 described homography matrixs of described tolerance demarcating module utilization form overdetermination homogeneous linear equations system, find the solution the inner parameter matrix K of the equation group acquisition picture pick-up device in the described equation system.

Images