CN101098492A - Method for determining drone target topological relation and camera calibration target capable of disposing arbitrarily - Google Patents

Abstract

The invention belongs to image test technique field, in particular to a method for confirming target topology relation and a camera mark target with free arrangement. The invention is characterized in that the camera mark target with free arrangement comprises four large rounds at opposite corners in a small round, while the centers of three large rounds are on one line, and the center of another large round is outside the line to be mark as the first round, and the large round nearest the first round is the third round, and the large round which center is at the middle of the three large rounds on one line is marked as zero round, and the left large round of the three ones is the second round, combined with image processing algorism and vector parallel judgment, the invention can easily find the topology relation of target with free arrangement. The invention is mainly used in image test.

Description

Determine the method for target topological relation and the camera calibration target that can put arbitrarily

Technical field

The invention belongs to computer vision technique, the image measurement technical field especially relates to the method for determining the target topological relation and the camera calibration target that can put arbitrarily.

Background technology

The design of target is the key technology of computer vision and image measurement technology, has determined the certainty of measurement and the simple and easy degree of system.Stereo target is made complexity, cost is high, and the Virtual Space target that generally adopts plane target drone to set up in the practical application is so the design of plane target drone also becomes the hot spot technology of this area research.The design key of plane target drone is the position relation of characteristic point on the plane picture of camera acquisition and the correspondence that the fact characteristic point position concerns, often is called as topological correspondence problem.

In the Tsai of video camera peg model, target generally is over against transducer, so use the spacing multiple principle of target, can determine that topology is corresponding.This target has produced variation because of projective transformation makes the spacing on the image planes when the video camera angle of inclination is big relatively, adopt spacing multiple principle to be easy to generate wrong topological relation.And plane target drone is in actual use, runs into target surface not over against the situation of transducer through regular meeting, for example demarcation of binocular camera system, and two video cameras will be seen target surface simultaneously, so target certainly can not be over against certain video camera wherein; In the camera calibration process of the uncertain visual angle of Zhang, require target in the sensor measurement visual field, to put the position of some different angles arbitrarily; In addition, when carrying out attitude measurement with plane target drone, the placement of target surface is also with the change change at random of object.So any putting position of necessary design one cover can both accurately be determined the target of topological corresponding relation.

Summary of the invention

For overcoming the deficiencies in the prior art, the object of the present invention is to provide a kind of method of definite target topological relation and the camera calibration target that can put arbitrarily, to realize putting arbitrarily the also accurate topological corresponding relation of extract minutiae of target.The technical solution used in the present invention is: a kind of camera calibration target that can put arbitrarily, the roundlet that comprises level and vertical arrangement constitutes, in the roundlet of arranging, be provided with and arrange 4 great circles that adopt the diagonal angle to place, the center of circle of three great circles is positioned at straight line, the center of circle of another great circle is beyond this straight line, the great circle label of the center of circle beyond this straight line is No. 1 circle, apart from No. 1 nearest great circle label of circle is No. 2 circles, the middle label of three great circles that the center of circle is positioned at straight line is No. 0 circle, and the remaining great circle label of three great circles that the center of circle is positioned at straight line is No. 3 circles.

A kind of method of definite target topological relation by the image of camera acquisition to the said demarcation target of claim 1, is carried out analyzing and processing to this image, determines the target topological relation, comprises the following steps:

(1) noise spot is removed in preliminary treatment, distinguishes great circle and roundlet

To the image that collects, carry out binaryzation, profile extraction respectively: at first, less circularity threshold value CT control is set, the influence that is used to remove the non-circular object of background; Then, remaining profile is divided into groups according to area and girth size, feature according to target, roundlet grouping is to comprise maximum that of profile certainly, remaining profile calculate respectively with roundlet grouping in each profile similarity and, four minimum profiles of its neutralization are defined as great circle, pick out all the other profiles;

(2) determine the target direction

For four great circles, three vectors are formed in the individual center of circle of the center of circle of arbitrary circle and its excess-three, if they do not have respectively a synteny, judge that then this arbitrary circle is No. 1 circle, any two circle center line connectings of its excess-three circle form a vector, calculate it respectively with the angle of No. 1 circle and three circle center line connecting vectors, size order according to angle can be determined circle respectively No. 2, No. 0 circle and No. 3 circles are the x axle with No. 0 circle and No. 1 round line, and its vertical direction is a Y-axis, determine coordinate system, with tense marker 0,1,2, the topological coordinate figure of No. 3 circles is respectively (0,0), (2,0), (1,1), (1 ,-1);

(3) seek x=0, x=1, y=0, the roundlet on four axles of y=-1

Travel through all roundlets, calculate the line and the X-axis vector in each center of circle and No. 0 center of circle

Synteny, the roundlet of conllinear is referred in the round sequence on the y=0 axle, simultaneously, utilize its line and No. 0 circle and No. 2 round heart lines Corner dimension differentiate be just/negative sense on, principle is: angle is acute angle (＜90 °), then forward, otherwise on negative sense; This circle sequence according to the distance ordering to No. 0 circle, is distributed corresponding topological coordinate figure respectively; The mark coordinate is that the roundlet of (1,0), (1 ,-1) and (0 ,-1) is respectively circle 4,5, No. 6.Finding label on y=0 axle circle sequence is No. 4 roundlet, travels through roundlet equally, calculates the synteny of the roundlet center of circle and No. 2 circle center line connectings and No. 4 roundlets and No. 2 circle center line connecting vectors, determines the round sequence of x=1 on axial; Same, find roundlet 5, No. 6, respectively by No. 3 great circle centers of circle and No. 5 roundlet circle center line connecting vectors,, determine the round sequence on y=-1 and x=0 are axially respectively by No. 0 great circle center of circle and No. 6 roundlet circle center line connecting vectors;

(4) determine the topological coordinate of other roundlets outside four

At first the roundlet topology coordinate that does not indicate is made as (0,0),, searches for the same roundlet B of x topology coordinate figure corresponding on the y=-1 axle, connect the vector in two centers of circle for the roundlet A on the y=0 axle Search for simultaneously all the other roundlet centers of circle and A circle center line connecting with Synteny, the y of the roundlet of conllinear topology coordinate figure is set as consistent with A, according to the round sequence on such method traversal y=0 axle; In like manner travel through the round sequence on the x=0 axle, utilize the point on the x=1 axle, determine the x topology coordinate figure of all the other roundlets;

(5) the topological coordinate figure that corrects mistakes

Processing through the 4th step, for horizontal stroke/vertical axis situation about causing different in size, the topological coordinate figure that roundlet is set one dimension may occur wherein and determine, and the error situation that another dimension is not set, at this situation, carry out following correction: for the x coordinate figure is the roundlet of O, can judge once more that the line of it and No. 0 circle is in the Y-axis vector

Conllinear if conllinear is not then rejected, in like manner, is 0 circle for the y coordinate figure, judges whether to be collinear in the x axial vector;

(6) reject unreliable circle

Because the influence of video camera projection conversion, the some of them circle may produce bigger distortion, becomes unreliable circle, by the threshold value of higher circularity, the whole circles after the numbering is further removed; Can also set suitable similarity threshold, use the similarity of other circles and No. 0 circle further to reject unreliable circle.

Said synteny is to use the sine value of vector angle to differentiate: the line segment vector

With The angle sine value can be expressed as:

$\sin \mathrm{\&theta;}=\sqrt{1-{\left(\frac{\stackrel{\&RightArrow;}{{\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="A20061001448100091.png"\; state="\{\{state\}\}">V</figure-callout>}}_1}\&CenterDot;\stackrel{\&RightArrow;}{V_d}}{\left|\right|\stackrel{\&RightArrow;}{{\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="A20061001448100091.png"\; state="\{\{state\}\}">V</figure-callout>}}_1}\left|\right|\&CenterDot;\left|\right|\stackrel{\&RightArrow;}{V_d}\left|\right|}\right)}^2}---\left(1\right)$

When this is worth less than threshold value sin θ _TThe time, can judge two line segment conllinear.

Said removal noise spot is to utilize the judgement of circularity to pick out the influence of part complex background to target, on image, for by N point { (x _i, y _j) | the closed outline that 1≤i≤n) surrounds, the area S of encirclement can be expressed as,

$S=\frac{1}{2}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}|x_{i-1}{y}_i-x_i{y}_{i-1}|---\left(2\right)$

I wherein, j is an integer, and it is r that the minimum of profile contains radius of a circle, and then circularity can be expressed as C=S/ π r ², the approaching more circle of its big more explanation profile of value, vice versa.

The present invention possesses following effect: because the present invention has adopted the target disc with directivity, the judgement parallel with vector of combining image Processing Algorithm is easy to the topological corresponding relation of the target of definite any putting position.The accuracy that topology is determined is high especially, and has the full-automation of handling, characteristics such as easy to use.

Description of drawings

Fig. 1 is a novel directivity plane target drone of the present invention.

Fig. 2 is the great circle wrong schematic diagram that binaryzation causes when adjacent.

Fig. 3 sets up schematic diagram for round label of the present invention and coordinate system.

Fig. 4 is that the use circularity of the present invention and the method for dividision into groups are rejected background noise profile schematic diagram.

The actually determined target directional diagram of Fig. 5.

Four of Fig. 6 round sequence chart on axially.

Fig. 7 target topology coordinate and error flag schematic diagram.

Fig. 8 target is handled example.

Embodiment

Further specify the present invention below in conjunction with drawings and Examples.

The present invention adopts one group of circular array as the target element, and wherein four great circles are determined the direction and the coordinate system of plane target drone, as shown in Figure 1.The arrangement of great circle adopts the diagonal angle to place, and has avoided in the binaryzation process, adjacent great circle zone is connected, as shown in Figure 2; Thereby can select the bigger circle of diameter, thereby the ratio that has guaranteed great circle and axis of small circle is relatively large, is easy to gather at a distance and handle.As shown in Figure 3, successively four circles in the target are carried out label, and stipulate to set up space coordinates according to shown in the figure, target has just had directivity like this.Therefore no matter target position how, as long as guaranteed that four label circles are complete, just can go out the direction of target according to following rule judgment, further determine correct topological corresponding relation.

1. core algorithm

(1) determines the core diagnostic method of target topological relation

Video camera has carried out the projection conversion to object, but area, girth ratio and synteny are the invariants in this conversion.So, at first illustrate the differentiation rule of the synteny of often using in the flow process below to propose to use the sine value of vector angle to differentiate here.For example, in Fig. 3, vector

With

The angle sine value can be expressed as:

$\sin \mathrm{\&theta;}=\sqrt{1-{\left(\frac{\stackrel{\&RightArrow;}{{\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="A20061001448100091.png"\; state="\{\{state\}\}">V</figure-callout>}}_1}\&CenterDot;\stackrel{\&RightArrow;}{V_d}}{\left|\right|\stackrel{\&RightArrow;}{{\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="A20061001448100091.png"\; state="\{\{state\}\}">V</figure-callout>}}_1}\left|\right|\&CenterDot;\left|\right|\stackrel{\&RightArrow;}{V_d}\left|\right|}\right)}^2}---\left(1\right)$

When this is worth less than threshold value sin θ _TThe time, can judge that two line segments are conllinear.In native system, the extraction deviation at the distortion of camera chain and profile center can produce the trace influence to synteny, thus can set here a very little threshold value (such as: sin3 °).

(2) core algorithm of place to go background noise

Utilize the judgement of circularity can pick out of the influence of part complex background to target.On image, for by N point { (x _i, y _j) | the closed outline that 1≤i≤N} surrounds, the area of encirclement can be expressed as,

$S=\frac{1}{2}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}|x_{i-1}{y}_i-x_i{y}_{i-1}|---\left(2\right)$

It is r that the minimum of profile contains radius of a circle, and then circularity can be expressed as C=S/ π r ², the approaching more circle of its big more explanation profile of value, vice versa.

The similarity differentiation is decided apart from invariant by Hu, and promptly figure is through translation and postrotational invariant.Suppose A, B is two profiles that will compare, and similarity has three definition,

$\left\{\begin{array}{c}{I}_1(A,B)=\underset{k=1}{\overset{7}{\mathrm{\&Sigma;}}}|\frac{1}{M_k^B}-\frac{1}{M_k^A}|\\ I_2(A,B)=\underset{k=1}{\overset{7}{\mathrm{\&Sigma;}}}|M_k^B-M_k^A|\\ I_3(A,B)=\underset{k}{\max }|(M_k^B-M_k^A)/M_k^A|\end{array}\right.---\left(3\right)$

Wherein, $M_k^A=\mathrm{sgn}\left(h_k^A\right){\log }_{10}\left|h_k^A\right|,$ $M_k^B=\mathrm{sgn}\left(h_k^B\right){\log }_{10}\left|h_k^B\right|,$ h _kBe k rank Hu distances.Can use one of them as discrimination principle, if two profiles are equal fully, I then ₁=I ₂=I ₃=0; Dissimilar more, three values are just more near 1.

2. determine the concrete steps of target topological relation

(1) noise spot is removed in preliminary treatment, distinguishes great circle and roundlet

To the image that collects, carry out binaryzation, profile extraction respectively.At first, less circularity threshold value C is set _TControl (for example 0.4), the influence that is used to remove the non-circular object of background: then, remaining profile is divided into groups according to area and girth size, according to the feature of target, the roundlet grouping is to comprise maximum that of profile certainly; Remaining profile calculate respectively with roundlet grouping in each profile similarity and, four minimum profiles of its neutralization are defined as great circle.Result as shown in Figure 4, the profile of unmarked centre coordinate is picked out.

(2) determine the target direction

For four great circles, three vectors are formed in the center of circle of the center of circle of arbitrary round A and its excess-three, if their do not have synteny respectively, then A must be No. 1 circle.Any two circle center line connectings of its excess-three circle form a vector, calculate it respectively with the angle of No. 1 circle and three circle center line connecting vectors, can determine circle respectively 2,0 and No. 3 according to the size order of angle.Coordinate system be can determine as shown in Figure 3, (0,0), (2,0), (1,1), (1 ,-1) are respectively with the topological coordinate figure of 0,1,2, No. 3 circle of tense marker.Actually look for the result as shown in Figure 5.

(3) seek x=0, x=1, y=0, the circle on four axles of y=-1

Travel through all roundlets, the line that calculates each center of circle and No. 0 center of circle with

Synteny, the roundlet of conllinear is referred in the round sequence on the y=0 axle, simultaneously, utilize its line and

Corner dimension differentiate be just/negative sense on, principle is: angle is acute angle (＜90 °), then forward, otherwise on negative sense; This circle sequence according to the distance ordering to No. 0 circle, is distributed corresponding topological coordinate figure respectively.On y=0 axle circle sequence, find circle No. 4, travel through roundlet equally, calculate the roundlet center of circle and No. 2 circle center line connectings and

Synteny, determine the round sequence on x=1 is axially; In like manner, can locate circle 5, No. 6, respectively by

With

, determine the round sequence on y=-1 and x=0 are axially.Search result as shown in Figure 6.

(4) determine the topological coordinate of other roundlets outside four

At first the roundlet topology coordinate that does not indicate is made as (0,0).For the round A on the y=0 axle, search for the same round B of x topology coordinate figure corresponding on the y=-1 axle, connect the vector in two centers of circle

Search for simultaneously all the other roundlet centers of circle and A circle center line connecting with Synteny, the y of the circle of conllinear topology coordinate figure is set as consistent with A, according to the round sequence on such method traversal y=0 axle; In like manner travel through the round sequence on the x=0 axle, utilize the point on the x=1 axle, determine the x topology coordinate figure of all the other circles.Fig. 7 is the example of handling in this step.

(5) the topological coordinate figure that corrects mistakes

Through the processing in the 4th step, for amount horizontal stroke/vertical axis situation about causing different in size, one dimension may appear wherein in the topological coordinate figure that all the other roundlets are set to be determined, and the error situation that another dimension is not set, the profile in the square frame as shown in Figure 7.At this situation, can carry out following correction: for the x coordinate figure is 0 circle, the line that can judge it and No. 0 circle once more be in Conllinear is not if conllinear is then rejected; In like manner, be 0 circle for the y coordinate figure, judge whether to be collinear in

(6) reject unreliable circle

Because the influence of video camera projection conversion, the some of them circle may produce bigger distortion, becomes unreliable circle.Can the whole circles after the numbering further be removed by the high threshold (for example more than 0.8) of circularity.In addition, can also set suitable similarity threshold (for example 0.08), similarities of using other circles and No. 0 circle are the unreliable circle of rejecting further.

At present, the present invention has successfully substituted the common plane target, is applied in the application such as camera calibration, binocular camera system calibrating and calculating target surface attitude of Zhang and Tsai.The present invention need in machine vision and the image measurement technology to be equally applicable to other occasions of plane target drone, has good using value.

Claims (3)

1. the camera calibration target that can put arbitrarily, the roundlet that comprises level and vertical arrangement constitutes, it is characterized in that, in the roundlet of arranging, be provided with and arrange 4 great circles that adopt the diagonal angle to place, the center of circle of three great circles is positioned at straight line, the center of circle of another great circle is beyond this straight line, the great circle label of the center of circle beyond this straight line is No. 1 circle, apart from No. 1 nearest great circle label of circle is No. 2 circles, the middle label of three great circles that the center of circle is positioned at straight line is No. 0 circle, and the remaining great circle label of three great circles that the center of circle is positioned at straight line is No. 3 circles.

2. the method for a definite target topological relation is characterized in that, by the image of camera acquisition to the said demarcation target of claim 1, this image is carried out analyzing and processing, determines the target topological relation, comprises the following steps:

(1) noise spot is removed in preliminary treatment, distinguishes great circle and roundlet

To the image that collects, carry out binaryzation, profile extraction respectively: at first, less circularity threshold value C is set _TControl, be used to remove the influence of the non-circular object of background, then, remaining profile is divided into groups according to area and girth size, according to the feature of target, roundlet grouping is to comprise maximum that of profile certainly, remaining profile calculate respectively with the roundlet grouping in each profile similarity and, four minimum profiles of its neutralization are defined as great circle, pick out all the other profiles;

(2) determine the target direction

For four great circles, three vectors are formed in the center of circle of the center of circle of arbitrary circle and its excess-three, if their do not have synteny respectively, judge that then this arbitrary circle is No. 1 circle; Any two circle center line connectings of its excess-three circle form a vector, calculate it respectively with the angle of No. 1 circle and three circle center line connecting vectors, can determine No. 2 circles, No. 0 circle and No. 3 circles respectively according to the size order of angle, be X-axis with No. 0 circle and No. 1 round line, and its vertical direction is a Y-axis, determine coordinate system, topological coordinate figure with 0,1,2, No. 3 circle of tense marker is respectively (0,0), (2,0), (1,1), (1 ,-1);

(3) seek x=0, x=1, y=0, the roundlet on four axles of y=-1

Travel through all roundlets, calculate the line and the X-axis vector in each center of circle and No. 0 center of circle

Synteny, the roundlet of conllinear is referred in the round sequence on the y=0 axle, simultaneously, utilize its line and No. 0 circle and No. 2 round heart lines

Corner dimension differentiate be just/negative sense on, principle is: angle is acute angle (＜90 °), then forward, otherwise on negative sense; This circle sequence according to the distance ordering to No. 0 circle, is distributed corresponding topological coordinate figure respectively; The mark coordinate is that the roundlet of (1,0), (1 ,-1) and (0 ,-1) is respectively circle 4,5, No. 6.Finding label on y=0 axle circle sequence is No. 4 roundlet, travels through roundlet equally, calculates the synteny of the roundlet center of circle and No. 2 circle center line connectings and No. 4 roundlets and No. 2 circle center line connecting vectors, determines the round sequence of x=1 on axial; Same, find roundlet 5, No. 6, respectively by No. 3 great circle centers of circle and No. 5 roundlet circle center line connecting vectors,, determine the round sequence on y=-1 and x=0 are axially respectively by No. 0 great circle center of circle and No. 6 roundlet circle center line connecting vectors;

(4) determine the topological coordinate of other roundlets outside four

At first the roundlet topology coordinate that does not indicate is made as (0,0),, searches for the same roundlet B of x topology coordinate figure corresponding on the y=-1 axle, connect the vector in two centers of circle for the roundlet A on the y=0 axle

Search for simultaneously all the other roundlet centers of circle and A circle center line connecting with

Synteny, the y of the roundlet of conllinear topology coordinate figure is set as consistent with A, according to the round sequence on such method traversal y=0 axle; In like manner travel through the round sequence on the x=0 axle, utilize the point on the x=1 axle, determine the x topology coordinate figure of all the other roundlets;

(5) the topological coordinate figure that corrects mistakes

Processing through the 4th step, for horizontal stroke/vertical axis situation about causing different in size, the topological coordinate figure that roundlet is set one dimension may occur wherein and determine, and the error situation that another dimension is not set, at this situation, carry out following correction: for the x coordinate figure is 0 roundlet, can judge once more that the line of it and No. 0 circle is in the Y-axis vector

Conllinear if conllinear is not then rejected, in like manner, is 0 circle for the y coordinate figure, judges whether to be collinear in the X-axis vector

(6) reject unreliable circle

Because the influence of video camera projection conversion, the some of them circle may produce bigger distortion, becomes unreliable circle, by the high threshold of circularity, the whole circles after the numbering is further removed; Can also set suitable similarity threshold, use the similarity of other circles and No. 0 circle further to reject unreliable circle.

3. the method for a kind of definite target topological relation according to claim 2 is characterized in that, said synteny uses the sine value of vector angle to differentiate the line segment vector

With

The angle sine value can be expressed as:

$\sin \mathrm{\&theta;}=\sqrt{1-{\left(\frac{\stackrel{\&RightArrow;}{V_1}\&CenterDot;\stackrel{\&RightArrow;}{V_d}}{\left|\right|\stackrel{\&RightArrow;}{V_1}\left|\right|\&CenterDot;\left|\right|\stackrel{\&RightArrow;}{V_d}\left|\right|}\right)}^2}$

When this is worth less than threshold value sin θ _TThe time, can judge two line segment conllinear.

Images