CN102034238B - Multi-camera system calibrating method based on optical imaging probe and visual graph structure - Google Patents

Abstract

The invention provides a multi-camera system calibrating method based on an optical imaging test head and a visual graph structure. The method comprises the following steps: independently calibrating each camera by the optical imaging test head to obtain the initial values of the internal parameter and aberration parameter of each camera; calibrating the multiple cameras two by two, and obtaining the fundamental matrix, polar constraint, rotation matrix and translation vector between every two cameras with a plurality of overlapped regions at a view field by means of linear estimation; building the connection relationship among the multiple cameras according to the graph theory and the visual graph structure, and estimating the rotation vector quantity initial value and translation vector quantity initial value of each camera relative to the referred cameras by a shortest path method; and optimally estimating all the internal parameters and external parameters of the all cameras and the acquired three-dimensional sign point set of the optical imaging test head by a sparse bundling and adjusting algorithm to obtain a high-precision calibrating result. The multi-camera system calibrating method is simple in a calibrating process from the partial situation to the overall situation and from the robust to the precise, ensures high-precise and robust calibration, and is applied to calibrating multi-camera systems with different measurement ranges and different distribution structures.

Description

Multi-camera system scaling method based on optical imagery gauge head and vision graph structure

Technical field

The invention belongs to technical field of visual measurement, relate to a kind of multi-camera system scaling method based on optical imagery gauge head and vision graph structure.

Background technology

Use multi-camera system to carry out three-dimensional measurement, must at first accomplish the intrinsic parameter and the distortion parameter of each video camera, the relative pose between multiple-camera reaches the demarcation with reference frame pose parameter.The measuring accuracy that robustness of demarcating and precision directly determine multi-camera system, thereby robust and high-precision demarcation are prerequisite and the conditions that multi-camera system is realized high-precision calibrating.

Camera calibration is of long duration, and the development through many decades has many efficient ways.In recent years, because of multi-camera system on large scale, large-range measuring unique obviously, receive efficient, the focus that high precision becomes research of extensive attention, the especially multi-camera system of domestic and international research.Yet in traditional multiple-camera scaling method, some adopt electronic theodolite to set up world coordinate system, thereby have increased the cost and the complicacy of system; Some adopt three-dimensional, two dimension, one dimension or 0 dimension to demarcate thing; Directly utilize the estimation of the position orientation relation completion multi-camera system calibrating parameters between single camera inner parameter and video camera; It is visible in all camera field of view to require to demarcate thing on the one hand; Calibration result can be influenced by noise, the error of calculation, initial parameter error etc. on the other hand, and precision is lower; Although certain methods has also adopted binding adjustment optimization method in the final stage of demarcating, yet the gatherer process of three-dimensional demarcation point set is comparatively complicated, and need abandon point not in the know, efficient is lower, and robustness is relatively poor.

Summary of the invention

The objective of the invention is to defective or deficiency, a kind of multi-camera system scaling method based on optical imagery gauge head and vision graph structure is provided to above-mentioned existing calibration technique.This method only needs an optical imagery gauge head of having demarcated; Just can accomplish the demarcation of demarcation of single camera inside and outside parameter and multi-camera system simultaneously; Method is simple; And do not need the optical imagery gauge head visible in all camera field of view, and only need two or interior visible the getting final product of above camera field of view therein, be applicable to various measure field.Adopt vision graph structure and binding adjustment to optimize,, guaranteed result's robustness and high precision by slightly demarcating to accomplish multi-camera system smartly.

For achieving the above object, the technical scheme that the present invention adopts is:

The foundation of step 1, hardware:

Adopt distributed a plurality of video camera and an optical imagery gauge head, all video cameras do not require overlapping space, visual field, but require that overlapped fov is arranged between per at least two video cameras:

Step 2, confirm the locus of each video camera; For each video camera is confirmed one or more space orientation points, by the principle of space orientation point selection promptly in the calibration process of single camera or multiple-camera the optical imagery gauge head all can be positioned in the camera field of view of being demarcated.In the demarcation of single camera; Reference for installation locating piece (1) on its corresponding anchor point at first; The optical imagery gauge head being inserted each camera field of view scope, and aim at benchmark locating piece (1), be center of circle rotation optical imagery gauge head with the survey of optical imagery gauge head point (2); Seven monumented points (5) on the optical imagery gauge head target body 4 are that the center of circle rotates with one heart to survey point 2, the multiple image of the different attitudes of camera acquisition optical imagery gauge head; Secondly, adopt Threshold Segmentation and ellipse fitting algorithm to extract the centre coordinate of each the LED sub-pixel in every width of cloth image, adopt gravity model appoach to calculate the centre coordinate of each monumented point 5; At last, according to the concentric circles rotation relationship of monumented point 5 and the space length invariant relation of having demarcated, recover the intrinsic parameter matrix and the distortion parameter of each video camera;

Step 3, reselect and locate the space orientation point, adjustment optical imagery gauge head attitude guarantees that the optical imagery gauge head is positioned at the overlapped fov of per two or above video camera; In overlapped fov; Rotary optical imaging gauge head; Obtain the LED image of optical imagery gauge head monumented point simultaneously by the overlapping video camera in visual field; Through Flame Image Process, center identification and monumented point coupling, and, recover fundamental matrix, essential matrix, polar curve geometric relationship and rotation matrix, translation vector between local two video cameras according to the geometrical constraint between monumented point;

Step 4, be node with each video camera; Per two video cameras to have overlapped fov are the limit; Make up the vision figure of multi-camera system, the direction on every limit is definite by the local rotation matrix and the translation vector of initial alignment, and by another video camera after the last camera points conversion; In vision figure; At first confirm by beginning, to set up the annexation between all video cameras with reference to video camera with reference to video camera; Adopt the rotation matrix and the translation vector of each video camera relative reference video camera of critical path method (CPM) calculating, realize the overall initial alignment of multi-camera system;

Step 5, according to the initial alignment result, the monumented point back projection that step 2 and step 3 are gathered, discerned, obtain demarcates point set to reference to camera coordinate system with this three-dimensional of setting up in the world coordinate system; Adopt sparse binding adjustment algorithm that whole calibrating parameters and the three-dimensional point set of demarcating are optimized estimation, obtain robust and high-precision calibration result;

Step 6, if calibration result has satisfied the measuring accuracy requirement, calibration process finishes; If do not satisfy; Then increase optical imagery gauge head anchor point quantity, images acquired is increased to original monumented point with the new three-dimensional calibration point of discerning and rebuild and concentrates again; Adopt sparse binding adjustment algorithm to optimize whole calibrating parameters and the three-dimensional point set of demarcating again, demarcate each parameter again.

Described each monumented point 5 is made up of six LED, and its center of gravity is the monumented point center.

Described step 2 comprises following steps:

Step 21, set up video camera pin hole projection model:

At first, set up world coordinate system X _WY _WZ _W, gauge head coordinate system X _tY _tZ _t, image coordinate yardstick coordinate system xy and pixel coordinate be uv, establishing has N=7 monumented point on the optical imagery gauge head,

Indicator sign point is coordinate in the gauge head coordinate system,

Indicator sign point is coordinate in world coordinate system, and (R is rotation and the translation parameter of video camera with respect to world coordinate system T), and K is the intrinsic parameter matrix of video camera, comprises 5 parameters, is respectively level and vertical direction focal length (f _x, f _y), image inclination factor s and photocentre coordinate (u0, v0), (d1 d2) is respectively radially distortion parameter with the tangential for k1, k2, and 7 monumented points (5) on the gauge head target surface (4) satisfy following projection relation:

${\mathrm{\λ}}^j{\tilde{x}}^j=\mathrm{KR}{X}_t^j+\mathrm{KT},j\∈7---\left(1\right)$

Wherein, λ ^jIt is the projection degree of depth of monumented point j; The normalization coordinate of indicator sign point j on image.

Step 22, uncalibrated image collection and mark point recognition: according to the principle of space orientation point selection; Select the space orientation point; And fixed reference locating piece (1); Is the optical imagery gauge head that rotate in the center of circle to survey point (2); Use image I 1 width of cloth of the different attitudes of camera acquisition; The image of gathering spreads all over the space, visual field of video camera as far as possible, adopts Threshold Segmentation and ellipse fitting algorithm to extract the sub-pixel center of each LED luminous point, adopts gravity model appoach to estimate the center of each index point;

Step 23, constant according to the concentric circles radius, when setting up each monumented point and rotating with the error function of surveying sharp distance, when the optical imagery gauge head when surveying point (2) and rotate, 7 monumented points are respectively with radius r _jRotate with one heart around surveying point (2), because all 7 concentrically ringed radius of gyration r are accurately demarcated in the position of monumented point _jConstant, constant according to the concentric circles radius, gauge head each monumented point in rotation process is constant to the distance of surveying point, and to rotating the I1 width of cloth image of gathering, each sign all can obtain the radius error equation:

${\mathrm{\Σ}}_{i=1}^{I1}(r_j^i-r_j)=0---\left(2\right)$

Wherein, Represent j (j=1 ..., 7) and the radius of gyration when the i time is rotated, estimated of individual monumented point, r _jRepresent j (j=1 ..., 7) and the calibration value of the individual monumented point radius of gyration;

The initial value of step 24, estimation camera interior and exterior parameter: because of the existence of error; Find the solution the initial value of camera intrinsic parameter matrix K, rotation matrix R and translation vector T through the formula of minimizing (3) radius error function, through back projection's error least estimated distortion parameter initial value

${\mathrm{\Σ}}_{j=1}^7{\mathrm{\Σ}}_{i=1}^{I1}(r_j^i-r_j)---\left(\text{3}\right)$

Step 25, the sparse binding adjustment algorithm of employing further improve stated accuracy.

Described step 3 comprises following steps:

Gauge head monumented point IMAQ in the step 31, overlapped fov: reselect and locate the space orientation point (L1, L2 ...); Adjustment optical imagery gauge head attitude; Guarantee that it is positioned at the overlapped fov of per two or above video camera; Rotary optical imaging gauge head obtains gauge head monumented point stereo-picture simultaneously by the overlapping video camera in visual field in this zone;

Step 32, stereo-picture mark point recognition and coupling: adopt Threshold Segmentation and ellipse fitting algorithm to extract the sub-pixel center of each LED luminous point, adopt gravity model appoach to estimate the center of each monumented point; According to monumented point position constraint and the constant rule of geometric moment, mate the right monumented point of stereo-picture fast;

Step 33, estimation fundamental matrix, polar curve geometric relationship:

Suppose that two video camera C1 and C2 have overlapping space, visual field, be without loss of generality that the initial point of establishing world coordinates is positioned at the center of video camera C1, the pose of video camera C1 is (R ₁, T ₁)=(I, 0), the pose of video camera C2 is (R ₂, T ₂), the relativeness of two video cameras be (R, T).Following relation is then arranged between the image coordinate

of the monumented point j of the same name that obtains of two video cameras:

${\mathrm{\λ}}_2^j{\tilde{x}}_2^j={\mathrm{\λ}}_1^j{K}_2{R}_2{K}_1^{-1}{\tilde{x}}_1^j+K_2{T}_2---\left(4\right)$

Estimate unknown scale parameter

and and obtain

that from this formula wherein

promptly is the fundamental matrix of requirement; Monumented point if any enough is right, 9 parameters of linear solution fundamental matrix.Simultaneously, in step 2, the intrinsic parameter matrix K of two video camera C1 and C2 ₁, K ₂Demarcate, use coordinate figure

The polar curve restriction relation of stereo-picture is estimated in (i=1,2):

$x_2^{\mathrm{jT}}{\mathrm{Ex}}_1^j=0---\left(5\right)$

Wherein,

is essential matrix, utilizes 7 algorithms can estimate essential matrix E;

Step 34, estimate relative pose between video camera and scale factor in twos:

Because the location aware of 7 monumented points on the gauge head; Can be unique confirm rotation matrix R and the translation vector T between per two video cameras; When estimating scale factor, normalization radius

and the actual radius of demarcating

that service marking point is surveyed the point rotation relatively can obtain:

${\mathrm{\λ}}^j=r^j/{\tilde{r}}^j---\left(6\right)$

Because the existence of error, the mean value of mean value or N measurement of getting 7 point estimation is as final scale factor λ;

$\mathrm{\λ}={\mathrm{\Σ}}_{j=1}^7{\mathrm{\λ}}^j/7---\left(7\right)$

$\mathrm{\λ}={\mathrm{\Σ}}_{i=1}^N({\mathrm{\Σ}}_{j=1}^7{\mathrm{\λ}}^j/7)/N---\left(8\right).$

Described step 4 comprises following steps:

Step 41, theoretical based on figure, the vision figure of structure multi-camera system:

With each video camera is node; Per two video cameras to have overlapped fov are the limit; Make up the vision figure of multi-camera system; In vision figure; The direction on every limit is temporarily definite by the local spin matrix and the translation vector of initial alignment, and by another video camera after the last camera points conversion;

Step 42, definite with reference to video camera rebulids the annexation between the video camera:

In vision figure, confirm one with reference to video camera, consider when selecting and the relation of adjacent camera with reference to video camera, by beginning, readjust and set up the annexation between the video camera with reference to video camera;

Step 43, adopt critical path method (CPM) to confirm that each video camera is with reference to rotation under the camera coordinate system and translational movement;

Use solving the shortest path from reference to the absolute position of video camera, establish C to each video camera _i, C _j, C _kBe the video camera that is coupled to each other on a certain path of vision figure, the video camera in twos to having demarcated obtains from C respectively _iTo C _jAnd C _jTo C _kTransformation matrix (R _Ij, T _Ij) and (R _Jk, T _Jk), from C _iTo C _kRelation by computes:

R _ik＝R _ijR _jk，T _ik＝T _ij+R _ijT _jk (9)

If from reference to unnecessary two of the node on the path of video camera, then repeatedly use following formula to find the solution, accomplish all video cameras with respect to absolute calibration with reference to video camera.

Described step 5 comprises following steps:

Step 51, obtain the initial three-dimensional point set of demarcating

By step 22 and step 32; Gather the also image coordinate of distinguishing mark point different calibration points in all video cameras; Whole calibrating parameters according to preceding 4 steps acquisition; Back projection obtains a sparse demarcation point set of three-dimensional; Because used the simple annexation of multiple-camera in demarcating, all there are error in the three-dimensional sparse demarcation point set and the calibrating parameters of this reconstruction;

Step 52, structure back projection error function

Suppose to have n three-dimensional coordinate point and m video camera, the projection coordinate on j camera review was x in i o'clock _Ij, the parameter of each video camera is by vectorial a _jExpression, each three-dimensional point availability vector b _iExpression, the projection of function Q () definition three-dimensional point on the video camera image planes, function d (x, the y) Euclidean distance between representative graph picture point x and the y according to the projective geometry relation, are set up following back projection's error function:

${\min }_{a_j,b_i}{\mathrm{\Σ}}_{i=1}^n{\mathrm{\Σ}}_{j=1}^{m}d{(Q(a_j,b_i),x_{\mathrm{ij}})}^2---\left(10\right)$

Step 53, use binding adjustment minimize back projection's error function, obtain accurate calibration result.

Described back projection error function is one and includes P ∈ ^MThe non-linear minimization problem of parameter vector, P ∈ R ^MParameter vector is by the pose parameter and the three-dimensional measurement point set X ∈ of all video cameras ^NForm.

The present invention is for realizing single camera inside and outside parameter and the relative pose demarcation of video camera in twos; Adopt the optical imagery gauge head as demarcating thing; 7 monumented points having demarcated the position are installed on the optical imagery gauge head, and (each monumented point is made up of 6 LED; Its center of gravity is the monumented point center), recover camera interior and exterior parameter according to monumented point on the gauge head target surface around the relation of rotatablely moving with one heart of surveying point, realize demarcating fast.Because measurement space is big; The visual field of video camera is limited; So the present invention regards multi-camera system as a vision graph structure relation, each video camera is the node of vision figure, and the limit of vision figure representes between node (video camera) overlapping field range is arranged; There is directivity on the limit of vision figure, represents pose to be transformed to another video camera by last video camera; On vision figure, optimize whole pose initial parameters of multi-camera system through shortest-path method.Simultaneously, because the error in noise and the coordinate transform can exert an influence to the entire system precision, the present invention further adopts sparse binding adjustment algorithm that the three-dimensional symbol point that obtains is optimized adjustment with whole initial alignment parameters, to improve the precision of calibrating parameters.Through using this method, the global error of system calibrating is reducing greatly, and the robustness of each parameter significantly improves.

Description of drawings

Fig. 1 is optical imagery gauge head and video camera projection relation synoptic diagram.

Label is wherein represented respectively: 1, benchmark locating piece, 2, survey point, 3, extension bar, 4, the gauge head target body, 5,7 monumented points (each monumented point is made up of 6 LED luminous points).X _WY _WZ _WRepresent world coordinate system, X _tY _tZ _tRepresent the gauge head coordinate system, xy image coordinate yardstick coordinate system, uv representative image pixel coordinate system; P1～P7 represents 7 monumented points on the optical imagery gauge head, and r1～r7 is respectively 7 monumented points to the distance of surveying point.

Fig. 2 has two video camera relative positions of overlapped fov to demarcate synoptic diagram.

Label wherein: C1～C8 represents 8 video cameras, and L1, L2 represent the position of anchor point, and R and T represent the relative pose transformation relation.

Fig. 3 makes up synoptic diagram for the vision figure of multi-camera system.

Label wherein: C1～C8 represents 8 video cameras.

Fig. 4 changes synoptic diagram for confirming with reference to the annexation of vision figure before and after the video camera.

Label wherein: C1～C8 represents 8 video cameras.

Fig. 5 is with reference to global calibration synoptic diagram under the camera coordinate system.

Label wherein: C1～C8 represents 8 video cameras, and L1, L2 represent the position of anchor point, and R and T represent the relative pose transformation relation.

Embodiment

Below in conjunction with accompanying drawing the present invention is described further.

Vision system to be made up of 8 video cameras is an example, specifies.The present invention uses optical imagery gauge head as shown in Figure 1.The optical imagery gauge head is mainly formed by surveying point (2), extension bar (3), target body (4), 7 monumented points (5) and LED luminous point thereof.The distribution of 8 video cameras is as shown in Figure 3, and all video camera does not require overlapping visual field, but per at least two video cameras have overlapping visual field.This method according to the space distribution of 8 video cameras, is selected and definite space orientation point earlier, utilizes the invariable rotary relation of optical imagery gauge head around the survey point, accomplishes the independent demarcation of single camera and the relative pose of per two video cameras and demarcates.Then, according to scheming the annexation that theory and vision figure make up multiple-camera, in demarcation with reference to completion multiple-camera under the camera coordinate system.For eliminating the initial alignment error, adopt sparse binding adjustment algorithm global optimization three-dimensional geometrical structure and whole calibrating parameters at last.

Below, by the demarcation order each step is specifically introduced.

Phase one: the independent of single camera intrinsic parameter and distortion parameter demarcated.

If video camera be numbered C1～C8, as shown in Figure 2.

Step 11, according to the optical imagery gauge head as shown in Figure 1 and the projection relation of video camera, set up world coordinate system X _WY _WZ _W, gauge head coordinate system X _tY _tZ _t, image coordinate yardstick coordinate system xy and pixel coordinate be uv.

Step 12, confirm the locus of each video camera; For each video camera is confirmed one or more space orientation points, by the principle of space orientation point selection promptly in the calibration process of single camera or multiple-camera the optical imagery gauge head all can be positioned in the camera field of view of being demarcated.In the demarcation of single camera; Fixed reference locating piece (1) on its corresponding anchor point at first; Is the optical imagery gauge head that rotate in the center of circle to survey point (2), uses image I 1 width of cloth of the different attitudes of camera acquisition, and the image of collection spreads all over the space, visual field of video camera as far as possible.Adopt Threshold Segmentation and ellipse fitting algorithm to extract the sub-pixel center of each LED luminous point; Adopt gravity model appoach to estimate centre coordinate (i=1 of each monumented point (forming) by 6 LED luminous points;, I1; J=1 ..., 7).Simultaneously, establish the coordinate of N=7 monumented point in the gauge head coordinate system on the optical imagery gauge head set up video camera according to formula (1) for for

coordinate in world coordinate system projection relation.

Step 13, according to the constant geometrical constraint of concentric circles radius, make up the radius error equation (2) of each monumented point.

Step 14, minimize the initial value that radius error function formula (3) recovers intrinsic parameter matrix K, rotation matrix R and the translation vector T of each video camera, through each distortion of camera parameter of back projection's error least estimated (k1, k2, d1, d2).

Step 15, adopt sparse binding adjustment algorithm, the inside and outside parameter of each video camera is optimized, improve stated accuracy.

Subordinate phase: have the relative pose of per two video cameras of overlapped fov to demarcate.

Get a pair of in 8 video cameras arbitrarily, by little numbering preceding, big numbering after sequential packet.

Step 21, as shown in Figure 2; Select and locate suitable space orientation point (L1, L2 ...); Adjustment optical imagery gauge head attitude; Guarantee that it is positioned at the overlapped fov of per two or above video camera, rotary optical imaging gauge head obtains gauge head monumented point stereo-picture simultaneously by the overlapping video camera in visual field in this zone.

Step 22, the sub-pixel center of adopting Threshold Segmentation and ellipse fitting algorithm to extract each LED luminous point adopt gravity model appoach to estimate the center of each monumented point.Based on index point position constraint and the constant rule of geometric moment, mate the right index point of stereo-picture fast.

Step 23, according to the formula (4) estimated that each of the camera unknown scale parameter

and

According relations

resume every two fundamental matrix of the camera; according to equation (5) to restore each of the camera epipolar constraint.

The rotation matrix R and the translation vector T of step 24, every pair of video camera of recovery calculate the yardstick factor lambda according to formula (7) and formula (8).

Phase III: based on the multi-camera system global calibration of vision figure.

Step 31, as shown in Figure 3 is a node with 8 video cameras, is the limit with per two video cameras with overlapped fov, makes up the vision figure of multi-camera system.In vision figure, the direction on every limit is temporarily confirmed by the local rotation matrix and the translation vector of initial alignment, is made up the directional vision figure shown in Fig. 4 (a).

Step 32, in vision figure, be with reference to video camera with C8, readjust and set up the annexation between the video camera, adopt critical path method (CPM) to confirm new directional vision figure, shown in Fig. 4 (b).

Step 33, according to formula (9), confirm each video camera with reference to rotation under the camera coordinate system and translational movement, as shown in Figure 5.Thus, tentatively accomplish all video cameras with respect to absolute calibration with reference to video camera.

Stage: the global optimization based on sparse binding adjustment is demarcated.

Step 41, by the initial alignment parameter, through the three-dimensional sparse demarcation point set X ∈ of backprojection reconstruction ^N

Step 42, set up back projection's error minimum function (10) by whole nominal datas and three-dimensional point set.

Step 43, use binding adjustment minimize back projection's error function, obtain accurate calibration result.

Five-stage: the further optimization of calibrating parameters is estimated, up to satisfying the requirement of three-dimensional measurement stated accuracy.

Characteristics of the present invention are:

This scaling method uses the imaging gauge head with fixed optics monumented point, carries out the demarcation of relative pose between camera interior and exterior parameter and video camera, and without any need for miscellaneous equipment, method is simple, and is with low cost.

This scaling method does not need the optical imagery gauge head visible in all camera field of view; Only need be it is thus clear that get final product in the overlapped fov of per two video cameras; Be applicable to the multi-camera system of different distributions structure, both can be used for the small scale space and demarcated, can be used for the large scale space again and demarcate.

Vision figure that this scaling method adopted and binding adjustment optimized Algorithm can significantly improve and demarcate efficient, robustness and precision.

Claims (4)

1. multi-camera system scaling method based on optical imagery gauge head and vision graph structure is characterized in that may further comprise the steps:

The foundation of step 1, hardware:

Adopt distributed a plurality of video camera and an optical imagery gauge head of having demarcated, all video cameras do not require overlapping space, visual field, but require between per at least two video cameras overlapped fov is arranged;

Step 2, confirm the locus of each video camera; For each video camera is confirmed one or more space orientation points, by the principle of space orientation point selection promptly in the calibration process of single camera or multiple-camera the optical imagery gauge head all can be positioned in the camera field of view of being demarcated, in the demarcation of single camera; Reference for installation locating piece (1) on its corresponding anchor point at first; The optical imagery gauge head being inserted each camera field of view scope, and aim at benchmark locating piece (1), be center of circle rotation optical imagery gauge head with the survey of optical imagery gauge head point (2); Seven monumented points (5) on the optical imagery gauge head target body (4) are that the center of circle rotates with one heart to survey point (2); The multiple image of the different attitudes of camera acquisition optical imagery gauge head, wherein each monumented point (5) is made up of six LED, and its center of gravity is the monumented point center; Secondly, adopt Threshold Segmentation and ellipse fitting algorithm to extract the centre coordinate of each the LED sub-pixel in every width of cloth image, adopt gravity model appoach to calculate the centre coordinate of each monumented point (5); At last, according to the concentric circles rotation relationship of monumented point (5) and the space length invariant relation of having demarcated, recover the intrinsic parameter matrix and the distortion parameter of each video camera;

Step 3, reselect and locate the space orientation point, adjustment optical imagery gauge head attitude guarantees that the optical imagery gauge head is positioned at the overlapped fov of per two or above video camera; In overlapped fov; Rotary optical imaging gauge head; Obtain the LED image of optical imagery gauge head monumented point simultaneously by the overlapping video camera in visual field; Through Flame Image Process, center identification and monumented point coupling, and, recover fundamental matrix, essential matrix, polar curve geometric relationship and rotation matrix, translation vector between local two video cameras according to the geometrical constraint between monumented point;

Step 4, be node with each video camera; Per two video cameras to have overlapped fov are the limit; Make up the vision figure of multi-camera system, the direction on every limit is definite by the local rotation matrix and the translation vector of initial alignment, and by another video camera after the last camera points conversion; In vision figure; At first confirm by beginning, to set up the annexation between all video cameras with reference to video camera with reference to video camera; Adopt the rotation matrix and the translation vector of each video camera relative reference video camera of critical path method (CPM) calculating, realize the overall initial alignment of multi-camera system;

Step 5, according to the overall initial alignment result of multi-camera system, the monumented point back projection that step 2 and step 3 are gathered, discerned, obtain demarcates point set to reference to camera coordinate system with this three-dimensional of setting up in the world coordinate system; Adopt sparse binding adjustment algorithm that whole calibrating parameters and the three-dimensional point set of demarcating are optimized estimation, obtain robust and high-precision calibration result;

Step 6, if calibration result has satisfied the measuring accuracy requirement, calibration process finishes; If do not satisfy; Then increase optical imagery gauge head anchor point quantity, images acquired is increased to original monumented point with the new three-dimensional calibration point of discerning and rebuild and concentrates again; Adopt sparse binding adjustment algorithm to optimize whole calibrating parameters and the three-dimensional point set of demarcating again, demarcate each parameter again.

2. the multi-camera system scaling method based on optical imagery gauge head and vision graph structure as claimed in claim 1, it is characterized in that: described step 2 comprises following steps:

Step 21, set up video camera pin hole projection model:

At first, set up world coordinate system X _WY _WZ _W, gauge head coordinate system X _tY _tZ _t, image coordinate yardstick coordinate system xy and pixel coordinate be uv, establishing has N=7 monumented point on the optical imagery gauge head,

Indicator sign point is coordinate in the gauge head coordinate system,

Indicator sign point is coordinate in world coordinate system, and (R is rotation and the translation parameter of video camera with respect to world coordinate system T), and K is the intrinsic parameter matrix of video camera, comprises 5 parameters, is respectively level and vertical direction focal length (f _x, f _y), image inclination factor s and photocentre coordinate (u0, v0), (d1 d2) is respectively radially distortion parameter with the tangential for k1, k2, and 7 monumented points (5) on the gauge head target surface (4) satisfy following projection relation:

${\mathrm{\λ}}^j{\tilde{x}}^j={\mathrm{KRX}}_t^j+\mathrm{KT},j=\mathrm{1,2},...,7---\left(1\right)$

Wherein, λ ^jIt is the projection degree of depth of monumented point j;

The normalization coordinate of indicator sign point j on image;

Step 22, uncalibrated image collection and mark point recognition: according to the principle of space orientation point selection; Select the space orientation point; And fixed reference locating piece (1); Is the optical imagery gauge head that rotate in the center of circle to survey point (2); Use image I 1 width of cloth of the different attitudes of camera acquisition; The image of gathering spreads all over the space, visual field of video camera as far as possible, adopts Threshold Segmentation and ellipse fitting algorithm to extract the sub-pixel center of each LED luminous point, adopts gravity model appoach to estimate the center of each index point;

Step 23, constant according to the concentric circles radius, when setting up each monumented point and rotating with the error function of surveying sharp distance, when the optical imagery gauge head when surveying point (2) and rotate, 7 monumented points are respectively with radius r _jRotate with one heart around surveying point (2), because all 7 concentrically ringed radius of gyration r are accurately demarcated in the position of monumented point _jConstant, constant according to the concentric circles radius, gauge head each monumented point in rotation process is constant to the distance of surveying point, and to rotating the I1 width of cloth image of gathering, each sign all can obtain the radius error equation:

${\mathrm{\Σ}}_{i=1}^{I1}(r_j^i-r_j)=0---\left(2\right)$

Wherein,

Represent j (j=1 ..., 7) and the radius of gyration when the i time is rotated, estimated of individual monumented point, r _jRepresent j (j=1 ..., 7) and the calibration value of the individual monumented point radius of gyration;

The initial value of step 24, estimation camera interior and exterior parameter: because of the existence of error; Find the solution the initial value of camera intrinsic parameter matrix K, rotation matrix R and translation vector T through the formula of minimizing (3) radius error function, through back projection's error least estimated distortion parameter initial value

${\mathrm{\Σ}}_{j=1}^7{\mathrm{\Σ}}_{i=1}^{I1}(r_j^i-r_j)---\left(3\right)$

Step 25, the sparse binding adjustment algorithm of employing further improve stated accuracy.

3. the multi-camera system scaling method based on optical imagery gauge head and vision graph structure as claimed in claim 1, it is characterized in that: described step 4 comprises following steps:

Step 41, theoretical based on figure, the vision figure of structure multi-camera system:

With each video camera is node; Per two video cameras to have overlapped fov are the limit; Make up the vision figure of multi-camera system; In vision figure; The direction on every limit is temporarily definite by the local spin matrix and the translation vector of initial alignment, and by another video camera after the last camera points conversion;

Step 42, definite with reference to video camera rebulids the annexation between the video camera:

In vision figure, confirm one with reference to video camera, consider when selecting and the relation of adjacent camera with reference to video camera, by beginning, readjust and set up the annexation between the video camera with reference to video camera;

Step 43, adopt critical path method (CPM) to confirm that each video camera is with reference to rotation under the camera coordinate system and translational movement;

Use solving the shortest path from reference to the absolute position of video camera, establish C to each video camera _i, C _j, C _kBe the video camera that is coupled to each other on a certain path of vision figure, the video camera in twos to having demarcated obtains from C respectively _iTo C _jAnd C _jTo C _kTransformation matrix (R _Ij, T _Ij) and (R _Jk,, T _Jk), from C _iTo C _kRelation by computes:

R _ik＝R _ijR _jk，T _ik＝T _ij+R _ijT _jk (9)

If from reference to the node on the path of video camera more than two, then repeatedly use following formula to find the solution, accomplish all video cameras with respect to absolute calibration with reference to video camera.

4. the multi-camera system scaling method based on optical imagery gauge head and vision graph structure as claimed in claim 2, it is characterized in that: described step 5 comprises following steps:

Step 51, obtain the initial three-dimensional point set of demarcating

By step 22; Gather the also image coordinate of distinguishing mark point different calibration points in all video cameras; Whole calibrating parameters according to preceding 4 steps acquisition; Back projection obtains a sparse demarcation point set of three-dimensional; Because used the simple annexation of multiple-camera in demarcating, all there are error in the three-dimensional sparse demarcation point set and the calibrating parameters of this reconstruction;

Step 52, structure back projection error function

Suppose to have n three-dimensional coordinate point and m video camera, the projection coordinate on j camera review was x in i o'clock _Ij, the parameter of each video camera is by vectorial a _jExpression, each three-dimensional point availability vector b _iExpression, the projection of function Q () definition three-dimensional point on the video camera image planes, function d (x, the y) Euclidean distance between representative graph picture point x and the y according to the projective geometry relation, are set up following back projection's error function:

${\min }_{a_j,b_i}{\mathrm{\Σ}}_{i=1}^n{\mathrm{\Σ}}_{j=1}^{m}d{(Q(a_j,b_i),x_{\mathrm{ij}})}^2---\left(10\right)$

Step 53, use binding adjustment minimize back projection's error function, obtain accurate calibration result.

Images