CN102385750B - Line matching method and line matching system on basis of geometrical relationship - Google Patents

Abstract

The invention discloses a line matching method and a line matching system on the basis of a geometrical relationship. The line matching method comprises the following steps of: a step 1 of carrying out line detection on a plurality of images; a step 2 of carrying out projection transformation on lines detected from the same image; a step 3 of obtaining an intersection point of circular arcs formed by the lines detected from the same image on a projection plane; a step 4 of generating a matching feature of the lines detected from the same image; and a step 5 of carrying out line matching by the matching features corresponding to the lines detected from each image. In the method, a fundamental matrix does not need to be obtained in advanced, the influence of the point matching precision on the precision of the line matching is avoided, and the efficiency of a matching algorithm is improved.

Description

Straight line matching process and system based on geometric relationship

Technical field

The present invention relates to image processing field, relate in particular to straight line matching process and system based on geometric relationship.

Background technology

Straight line is to constitute important geometric element in the object in the three-dimensional world.The linear feature of object still keeps in the two dimensional image that obtains from three-dimensional scenic.In the image of under the different visual angles Same Scene being taken a part of straight line remain unchanged or straight line between have certain correlativity, find the straight line of the correspondence in the different images, utilize the corresponding relation between the straight line to determine that coupling and transformation relation between two width of cloth images have important meaning.The recovery of three-dimensional scenic mainly utilizes a little and line comes out the reproductions such as object target of three-dimensional, and the straight line in the coupling different images is the important foundation of three-dimensional reconstruction.

The method of straight line coupling mainly adopts utmost point geometrical constraint to carry out coupling just, and the recycling correlation method carries out the essence coupling.Utmost point geometrical constraint method is to utilize the geometric relationship of the image that obtains under two visual angles, set up the corresponding geometric transformation relation between picture point, the geometrical constraint of point in the image at another visual angle in piece image can be represented by fundamental matrix, the fundamental matrix that obtains between image utilizes two end points of straight line can obtain the zone of meeting geometric constraint in another visual angle image, belong to the regional straight line of geometrical constraint as candidate's straight line, finish thick coupling.Smart coupling adopts relevant method, and the grey scale change of choosing pixel in the straight line neighborhood realizes coupling as similarity.

Adopt the method for utmost point geometrical constraint need ask for fundamental matrix in advance, asking for of fundamental matrix is to adopt match point that the method for asking linear equation is realized, before the straight line coupling, need to carry out a coupling, the precision receptor site matching precision influence of straight line coupling, matching algorithm efficiency is not high.

Summary of the invention

At the above-mentioned problems in the prior art, the invention provides straight line matching process and system based on geometric relationship.

The invention provides the straight line matching process based on geometric relationship, comprising:

Step 1 is carried out straight-line detection to several images;

Step 2, to from same image detection to straight line carry out projective transformation;

Step 3, ask for same image detection to the intersection point of the circular arc that forms on the projecting plane of straight line;

Step 4 generates the matching characteristic of the straight line that same image detection arrives;

Step 5, utilize from each image detection to the matching characteristic of straight line correspondence carry out the straight line coupling.

In one example, in the step 1, thereby edge of image detected the straight line that detects in the image.

In one example, in the step 2, with image

Geometric center as the central point of coordinate transform, projection centre is the central point of unit sphere, wherein L ₁, L ₂, L ₃..., L _nBe detected straight line in the same image.

In one example, in the step 2, establishing the coordinate of coordinate transform central point in image is (q _X0q _Y0), in the same image on the detected straight line coordinate of any point be (q _xq _y), then The Transformation Relation of Projection is:

$\frac{\left(\begin{array}{ccc}{q}_x-q_{x0}& q_y-q_{y0}& -1\end{array}\right)}{\left|\right|\left(\begin{array}{ccc}{q}_x-q_{x0}& q_y-q_{y0}& -1\end{array}\right)\left|\right|}.$

In one example, in the step 3, establish on first circular arc of detected two straight line correspondences in the same image and 2 coordinate vector on second circular arc is respectively

With

Then the intersection point of circular arc is

$\±\frac{({\mathrm{CP}}_{11}^1\×{\mathrm{CP}}_{12}^1)\×({\mathrm{CP}}_{21}^1\×{\mathrm{CP}}_{22}^1)}{\left|\right|\left({\mathrm{CP}}_{11}^1{\×\mathrm{CP}}_{12}^1\right)\×({\mathrm{CP}}_{21}^1\×{\mathrm{CP}}_{22}^1)\left|\right|}.$

In one example, step 4 comprises:

Step 41 is chosen the corresponding distance between two points on straight line of the intersection point coordinate points p farthest of camber line _Max1, p _Max2, and with its mid point As new coordinate transform central point, and detected straight line in the same image carried out new projective transformation;

Be coordinate points p _Max1, p _Max2At the mid point of X-axis,

Be coordinate points p _Max1, p _Max2Mid point in Y-axis;

Step 42 is asked for coordinate points p _Max1, p _Max2The unit normal vector of place circular arc under new projection, and this unit normal vector is set to reference vector;

Step 43, the intersection point that calculates detected straight line in the same image under new projection coordinate vector and the angle of reference vector;

Step 44 sorts the back as the matching characteristic of straight line with angle.

In one example, in the step 5, utilize nearest criterion that the matching characteristic of straight line is mated.

The invention provides a kind of system that realizes based on the straight line matching process of geometric relationship.

The present invention need not to ask in advance fundamental matrix, has avoided the influence of some matching precision to the precision of straight line coupling, has improved the efficient of matching algorithm.

Description of drawings

Come the present invention is described in further detail below in conjunction with accompanying drawing, wherein:

Fig. 1 is the straight line matching process process flow diagram based on geometric relationship of the present invention;

Fig. 2 is the projective transformation illustraton of model;

Fig. 3 is that the circular arc intersection point is asked for schematic diagram.

Embodiment

The present invention proposes a kind of new straight line matching process, utilize the geometric relationship between the image cathetus to carry out matching judgment.Have parallel and overlapping relation between straight line in the image and the straight line, crossing or parallel relation between the image straight line that obtains under the different visual angles is constant under certain condition, determine the geometric relationship of straight line and other straight lines from straight line, utilize whether a straightforward differentiation of geometric relationship is same straight line.After linear projection is to the sphere, there are two intersection points in per two straight lines at sphere, this two intersection point is two limits of sphere great circle, choose limit as the expression amount of geometric relationship, with the geometric relationship employing vector representation of each bar straight line with other straight lines, utilize vector correlation to mate as matching criterior.

Method provided by the invention as shown in Figure 1, it has used 2 images (being not limited to 2 images), i.e. image 1 and image 2, but identical to each treatment of picture flow process.Method provided by the invention comprises:

Step 101, image 1 and image 2 are carried out straight-line detection.When carrying out straight-line detection, by to Image Edge-Detection, and then from the image border, detect straight line.Former two width of cloth images are designated as respectively: f ¹(x y) and f ²(x y) detects image f ¹Straight line n bar in (x y) is designated as respectively

Image f ²Straight line m bar in (x y) is designated as respectively

Belong on the straight line $\begin{array}{cc}{L}_i^k& \left(\begin{array}{c}k=1,i=\mathrm{1,2},...n\\ k=2,i=\mathrm{1,2},...m\end{array}\right)\end{array}$ On j the point be expressed as

Only there is the image of rectilinear point to be designated as ∏ ₁, ∏ ₂, be expressed as

Step 102, projective transformation.Be on the same orthodrome on the projecting plane through the point on the same straight line on the projective transformation image.Certainly exist two intersection points between different circular arcs, and the line of two intersection points passes through the centre of sphere.The position of the intersection point that the geometry of image cathetus and relation are different corresponding is just different, for example in the image two straight line parallels then the circular arc intersection point on the sphere is near the equator of projection model, the circular arc intersection point that two straight lines intersect in image on the sphere then departs from the equator.Position of intersecting point information response between spherical arc the geometric relationship between straight line.The projective transformation model as shown in Figure 2, with image f ¹Straight line in (x y)

The geometric relationship proper vector is asked for the process into example explanation projective transformation. Be image ∏ ₁In straight line, choose ∏ ₁Geometric center

Be the central point of coordinate transform, projection centre C elects the center of sphere as, chooses unit sphere, connects point on the straight line

C gets projection line with projection centre, the intersection point of projection line and sphere

For Subpoint on sphere is with ∏ ₁On point adopt The Transformation Relation of Projection to project on the sphere.Note: image ∏ ₁In Value is at 1,

The coordinate of coordinate transform central point in image is

Subpoint cp then ¹Coordinate is:

$\frac{\left(\begin{array}{ccc}{q}_x^1-q_{x0}^1& q_y^1-q_{y0}^1& -1\end{array}\right)}{\left|\right|\left(\begin{array}{ccc}{q}_x^1-q_{x0}^1& q_y^1-q_{y0}^1& -1\end{array}\right)\left|\right|}$

Step 103, ask for how much intersection points.Same orthodrome and projection centre can be determined unique plane, and the line that intersects the intersection point of circular arc in twos was the intersection of two Plane intersects of circular arc and projection centre.Can obtain the direction vector of intersection by two normal vectors that intersect the plane.Intersection is crossed the centre of sphere as previously shown, and intersection point is positioned on the sphere, and the coordinate of intersection point can be obtained by the direction vector of intersection.With straight line

And straight line

Be example, image ∏ ₁Cathetus

Last 2 subpoints on sphere

With

Coordinate vector be

Straight line

Last 2 subpoints on sphere

With

Coordinate vector Then

The projection circular arc with The intersection point of projection circular arc is:

Two intersection points are about centre of sphere symmetry, and amount of orientation first component is not less than zero value and is designated as Obtain other circular arcs together

The intersection point of projection circular arc,

At ∏ ₁On corresponding point

In like manner ask for the intersection point between each circular arc successively, i.e. image ∏ ₁Cathetus With

The intersection point of projection circular arc is

Corresponding point in original image are

Step 104, generation matching characteristic.In this step, utilize the inverse transformation of projective transformation to obtain the picture point of circular arc intersection point correspondence, these points are the straight-line intersection in the image.Choose the straight-line intersection middle distance farthest 2 mid point as new transform center all intersection points are projected on the sphere again with new transform center, distance farthest the unit normal vector of the circular arc at 2 places as with reference to vector.The angle (or inner product absolute value) of the coordinate vector of intersection point and reference vector is as the expression amount of intersection point under new projection, and this expression amount is as an element of linear feature.To belong to the expression amount of intersection point on the same straight line by arranging the generating feature vector from small to large.

As example, select straight-line intersection middle distance 2 points farthest,

2 mid points:

$p_{\mathrm{mid}}^1:\begin{array}{c}\left(\begin{array}{cc}{p}_{{\mathrm{mid}}_x}^1& q_{{\mathrm{mid}}_y}^1\end{array}\right)=\end{array}\left(\begin{array}{cc}\frac{p_{{\max 1}_x}^1+p_{{\max 2}_x}^1}{2}& \frac{p_{{\max 1}_y}^1+p_{\max 2y}^1}{2}\end{array}\right).$ With

For transform center with intersection point

Project to the projection coordinate that looks for novelty on the sphere.

Straight-line intersection

Projection coordinate's vector is under new projection:

Account for

Coordinate vector at the subpoint of newly casting:

The unit normal vector of crossing these 2 circular arcs is: $\stackrel{\‾}{C}{V}_{\max }^1=\frac{\stackrel{\‾}{C}{p}_{\max 1}^1\×\stackrel{\_}{C}{p}_{\max 1}^1}{\left|\right|\stackrel{\‾}{C}{p}_{\max 1}^1\×\stackrel{\‾}{C}{p}_{\max 1}^1\left|\right|},$ Note

Be reference vector

Intersection point projection coordinate vector And reference vector

Inner product:

Straight line

Proper vector generates:

Go up each intersection point projection coordinate vector

Each intersection point projection coordinate vector is with the angle of reference vector:

$\mathrm{ar}\cos (\stackrel{\‾}{C}{V}_{\max }^1\·[\stackrel{\‾}{C}{P}_{12}^1;\stackrel{\‾}{C}{P}_{13}^1;......;\stackrel{\‾}{C}{P}_{1n-1}^1{]}^T).$

Right Sort by size the result

$\mathrm{sort}\left(\mathrm{ar}\cos (\stackrel{\‾}{C}{V}_{\max }^1\·[\stackrel{\‾}{C}{P}_{12}^1;\stackrel{\‾}{C}{P}_{13}^1;......;\stackrel{\‾}{C}{P}_{1n-1}^1{]}^T)\right)$

Then matching characteristic is expressed as:

${\mathrm{MG}}_1^1=\mathrm{sort}\left(\mathrm{ar}\cos (\stackrel{\‾}{C}{V}_{\max }^1\·[\stackrel{\‾}{C}{P}_{12}^1;\stackrel{\‾}{C}{P}_{13}^1;\·\·\·\·\·\·;\stackrel{\‾}{C}{P}_{1n-1}^1{]}^T)\right).$

Generate the geometric relationship matching characteristic of all straight lines,

Adopt same step to generate image ∏ ₂Feature

Step 105, straight line coupling.Utilize the proper vector of nearest criteria match straight line.The above only is preferred implementation of the present invention, but protection domain of the present invention is not limited thereto.Any those skilled in the art all can carry out suitable change or variation to it in technical scope disclosed by the invention, and this change or variation all should be encompassed within protection scope of the present invention.

Claims (3)

1. based on the straight line matching process of geometric relationship, it is characterized in that, comprising:

Step 1 is carried out straight-line detection to several images;

Step 2, to from same image detection to straight line carry out projective transformation, wherein, with image

Geometric center as the central point of coordinate transform, projection centre is the central point of unit sphere, wherein L ₁, L ₂, L ₃..., L _nBe detected straight line in the same image, establishing the coordinate of coordinate transform central point in image is (q _X0q _Y0), in the same image on the detected straight line coordinate of any point be (q _xq _y), then The Transformation Relation of Projection is:

Step 3, ask for same image detection to the intersection point of the circular arc that forms on the projecting plane of straight line, establish on first circular arc of detected two straight line correspondences in the same image and 2 coordinate vector on second circular arc is respectively

With

Then the intersection point of circular arc is

Step 4 generates the matching characteristic of the straight line that same image detection arrives, and is in described step 4, further comprising the steps of:

Step 41 is chosen the corresponding distance between two points on straight line of the intersection point coordinate points p farthest of camber line _Max1, p _Max2, and with described coordinate points p farthest _Max1, p _Max2Mid point

As new coordinate transform central point, and detected straight line in the same image carried out new projective transformation; Be coordinate points p _Max1, p _Max2At the mid point of X-axis,

Be coordinate points p _Max1, p _Max2Mid point in Y-axis;

Step 42 is asked for coordinate points p _Max1, p _Max2The unit normal vector of place circular arc under new projection, and this unit normal vector is set to reference vector;

Step 43, calculate detected straight-line intersection in the same image under new projection coordinate vector and the angle of reference vector;

Step 44 sorts the back as the matching characteristic of straight line with angle;

Step 5, utilize from each image detection to the matching characteristic of straight line correspondence carry out the straight line coupling.

2. the straight line matching process based on geometric relationship as claimed in claim 1 is characterized in that, in the step 1, thereby edge of image is detected the straight line that detects in the image.

3. the straight line matching process based on geometric relationship as claimed in claim 1 or 2 is characterized in that, in the step 5, utilizes nearest criterion that the matching characteristic of straight line is mated.

Images