CN105957074A - Line segment matching method and system based on V-shape intersection description and local homography matrix - Google Patents

Abstract

The invention provides a line segment matching method and a system based on V-shape intersection description and local homography matrix. The matching method comprises the following steps: constructing V-type intersections, and describing the characteristics of the V-type intersection, matching the V-shape intersections through the calculations of the similarities among the vectors of the described characteristics, eliminating erroneous matching through the use of the principle to make adjacent V-type intersections' topological distribution maintain relative stability and increasing the chances of right matching; conducting single line segment matching, including the line segments in the pairs that are matched not through the formation of V-shaped intersections, and based on the estimation of a local homography matrix, obtaining the final matching result of the line segments. The invention can achieve good effects under various image transformation conditions, like rotation, scaling and angle changing. Compared with the existing method, the technical scheme provided by the invention can find more matches and keep higher matching precision in most cases.

Description

Describe based on V-type intersection point and the line match method and system of local homography matrix

Technical field

The invention belongs to digital image processing techniques field, particularly relating to a kind of description based on line segment V-type intersection point should with local list The line match technical scheme of Matrix Estimation.

Background technology

Line segment can represent the profile of scene very well, and especially artificial scene, such as building, indoor etc.；A small amount of edge line segment Just can comprise the profile information of whole scene.Thus, relative to method based on image characteristic point, based on line Segment feature The method of the information obtaining scene captured by image has certain advantage.Want to utilize line Segment feature to carry out obtaining field The information of scape, a very important step be intended to by line match obtain image between line segment of the same name.Because only that obtain Take image between line segment of the same name, could be by the different images contact of shooting Same Scene under different visual angles, varying environment Get up, thus utilize plurality of pictures to obtain that scene is the most complete and accurate information.

Existing line match method can be divided into following two generally: line segment becomes single matching process and the match party in groups of line segment Method.In the one-tenth list matching process of line segment, can be divided into again based on line segment regional area describe method and based on Point matching auxiliary Method.In line segment in groups matching process, line segment knot group's matching process and the peering matching process of line segment can be subdivided into again.

Based on the regional area at line segment place is carried out this kind of method of feature description by utilize line segment peripheral region pixel gray scale, Gradient, color form the feature description to line segment, realize the coupling of line segment by weighing the similarity between arc description.This Class method is based on such a common hypothesis: there is bigger overlapping region between line segment of the same name.Thus between two line segments of the same name Overlapping region less, during even without overlapping region, this method tends not to enough realize the successful match of this two lines section. It addition, repeating texture region, this method easily produces erroneous matching.This is because the line descriptor for a segment in this region Difference the least, be difficult to find and most preferably mate line segment by weighing the similarity described between son.

First this kind of line match method based on Point matching auxiliary finds Point matching between image, then utilizes three-dimensional coplanar dotted line Between the affine or projective invariant that exists evaluate the similarity between line segment.If the two lines section from two images meets These invariants, being considered as this two lines section is line segment of the same name.What the advantage of this kind of method was that they utilize is between dotted line Geometrical relationship rather than the local features of line segment, even if thus occurring relatively along with image conversion when line segment place regional area During big deformation, it is also possible to obtain reasonable matching effect.But, this kind of method be jointly disadvantageous in that line match Result depends critically upon Point matching result, when abundant Point matching cannot be obtained, as when image scene texture rareness, this The method of kind is difficult to obtain preferable effect.

Adjacent segments in the line segment that first the peering matching process of line segment will extract from two pictures intersects, and forms intersection point；So Afterwards by mating these intersection points formed, obtain line match.This kind of method is the most feasible to be because once from two pictures An antinode successful match, formed this antinode two pairs of line segments also by Auto-matching.Line match is turned by this type of method Turn to this most ripe problem of Point matching, thus the Point matching method of some maturations can be utilized to obtain line match.So And, the intersection point of the line segment of two is not only the characteristic point obtained by some feature point extraction operator, it comprise formed it two The point of bar line segment-line relation information.How to effectively utilize these information to assist intersection point to mate, obtain the intersection point of more robust Method of completing the square, also will be one is worth studying a question.

A kind of last line match method is that line segment ties group's matching process.First this kind of method closes on distribution relation and line according to space The line segment extracting from two images is divided into different lines by the significance of Duan Zishen (on line segment shade of gray meansigma methods) a little Stage group, obtains line segment group coupling, the then matching relationship of the line segment in determining coupling group one by one by mating these line segments group. During the matching relationship finding line segment group, the method converts at image according to the spatial distribution between one group of line segment closed on During keep relative stability such a thought, by calculating in line segment group the relative position relation between line segment, obtain line segment Description of group.By comparing the sub-similarity of description, method obtains line segment group coupling.This method is corresponding at line segment end points of the same name When relation is not accurate, often produces erroneous matching or miss correct coupling therein.

Summary of the invention

In order to solve problem present in above method, the image at Various Complex converts and obtains more line match in scene, The invention provides a kind of description and the line match technical scheme of local homography matrix estimation based on line segment V-type intersection point.

The present invention provides a kind of line match method estimated based on V-type intersection point and local homography matrix, if to be matched two figures As respectively I and I ', from I and I ', extract line segment respectively, if the line segment aggregate extracted is respectivelyWith Wherein M and N is illustrated respectively in the total number of line segment of the middle extraction of I and I ', is found by following steps and is present inWithIn line segment of the same name,

Step 1, V-type intersection point builds, is included in the set respectively obtaining V-type intersection point in image I and I 'With The sum of the V-type intersection point formed during wherein P and Q is illustrated respectively in two images；

Step 2, V-type intersection point mates, including in setWithIn V-type intersection point carry out feature description, obtain V-type intersection point Feature description vector, by calculating the similarity between feature description vector, V-type intersection point is mated, if coupling obtains Q The set that individual V-type intersection point is formed

Step 3, V-type intersection point matching optimization, including to step 2 gainedUtilization closes on V-type intersection topology The principle that distribution keeps relative stability rejects erroneous matching, and increases correct coupling；

Step 4, single line match, including rightWithIn not over formed V-type intersection point carry out The line segment of coupling, estimates to carry out matching treatment, gained line match result and step 3 gained line segment based on local homography matrix Join result and constitute final line match result together.

And, in step 3, described in close on V-type intersection topology distribution keep relative stability, it is judged that mode is as follows,

If V-type intersection point j_cA coordinate system is constituted, with j with the two lines section and line segment extended line forming it_cAdjacent is the most matched V-type intersection point be distributed in four quadrants of this coordinate system, when meeting following condition, it is judged that close on V-type intersection topology distribution Keep relative stability,

After image converts, work as j_cBecome j '_c, with j '_cThe most matched V-type intersection point closed on be j_cNeighbours same place surpass Cross preset ratio, and a pair same place of any of which, a little at j_cQuadrant in the coordinate system formed and another point are at j '_cInstitute's shape The quadrant of the coordinate system become keeps identical probability to exceed predetermined probabilities.

And, in step 3, described rejecting erroneous matching, and increase correct coupling, it is achieved it is as follows,

Step 3.1, initial threshold value d₁、d₂Initial value, current iteration frequency n=1；

Step 3.2, deletesPresent in erroneous matching, anyIt is unsatisfactory for following two conditions will be deleted: j_qWith j '_qMeeting the constraint of core line, they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

j′_qWith j '_qHave and close on the distribution of V-type intersection topology and keep relative stability this characteristic；

Step 3.3, increases in the V-type intersection point not mated and match from two imageIn, it is assumed that j and j ' is to come respectively for a pair From the V-type intersection point not mated of two pictures, and if only if they when meeting following three condition, just think that (j, j ') is correct Join:

J and j ' meets the constraint of core line, and they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

The distance between feature description vector corresponding for j and j ' is less than certain predetermined threshold value d₂；

J and j ' has and closes on the distribution of V-type intersection topology and keep relative stability this characteristic；

Step 3.4, it may be judged whether meet the iteration termination condition preset, meets and terminates iteration, enters step 4, otherwise makes n=n+1, Make d₁=d₁+ c, returns step 3.2 and continues executing with, and wherein, c is default step-length.

And, described iteration termination condition is, the matching number in current collection and last iteration perform after step 3.3 Join number to compare and do not increase, or current iteration number of times is more than preset times threshold value.

And, in step 4, described based on local homography matrix estimate carry out matching treatment, it is achieved as follows,

Step 4.1, carries out local homography matrix and estimates, including the constraint utilizing the V-type intersection point of a pair coupling to be provided, it is right to obtain Should be in this V-type intersection point homography matrix to the local matching region at place；

Step 4.2, mates single line segment, it is achieved as follows,

IfBy being mated, by K the V-type intersection point obtained by above-mentioned steps 3, the set formed,WithIt is illustrated respectively in the remaining line segment aggregate also not matched in two width figures；For each line SectionOrFind the V-type intersection point that closest 4 have matched, and this line segment is assigned to accordingly In group；

Assuming that l and l ' is a pair line segment to be matched, be respectively allocated in the V-type intersection point j and j ' of coupling, wherein j is by line segment pair (l_m,l_n) formed, j ' by line segment to (l '_m,l′_n) formed, first determine whether the angle difference of l and l ' whether with line match (l_m,l′_m) and (l_n,l′_n) angle difference of middle conductor is similar；

If it is by step 4.1 gained from the corresponding homography matrix H of the V-type intersection point (j, j ') of coupling, (l, l ') is evaluated, It is mapped in the plane of delineation at l ' place by H including by the two of l intersection points, obtains a new line segment l_hIf, l_hTwo ends Point is to the distance of l 'WithSimultaneously less than respective threshold d_t, it is believed that (l, l ') is a pair correct line match.

The present invention provides a kind of line match system estimated based on V-type intersection point and local homography matrix, by arranging with lower module Find and be present inWithIn line segment of the same name,

V-type intersection point builds module, is respectively I and I ' for setting to be matched two image, extracts line segment from I and I ' respectively, if The line segment aggregate extracted is respectivelyWithWherein M and N is illustrated respectively in the middle extraction of I and I ' The total number of line segment, respectively obtains the set of V-type intersection point in image I and I 'WithWherein P and Q is illustrated respectively in two images the sum of the V-type intersection point formed；

V-type intersection point matching module, is used for including in setWithIn V-type intersection point carry out feature description, obtain V-type hand over The feature description vector of point, mates V-type intersection point by calculating the similarity between feature description vector, if coupling obtains The set that Q V-type intersection point is formed

V-type intersection point matching optimization module, for V-type intersection point matching module gainedUtilization closes on V-type The principle that intersection topology distribution keeps relative stability rejects erroneous matching, and increases correct coupling；

Single line match module, it is right to be used for includingWithIn not over formed V-type intersection point enter The line segment of row coupling, estimates to carry out matching treatment, gained line match result and V-type intersection point coupling based on local homography matrix excellent Change module gained line match result and constitute final line match result together.

And, in V-type intersection point matching optimization module, described in close on V-type intersection topology distribution keep relative stability, it is judged that mode It is as follows,

If V-type intersection point j_cA coordinate system is constituted, with j with the two lines section and line segment extended line forming it_cAdjacent is the most matched V-type intersection point be distributed in four quadrants of this coordinate system, when meeting following condition, it is judged that close on V-type intersection topology distribution Keep relative stability,

After image converts, work as j_cBecome j '_c, with j '_cThe most matched V-type intersection point closed on be j_cNeighbours same place surpass Cross preset ratio, and a pair same place of any of which, a little at j_cQuadrant in the coordinate system formed and another point are at j '_cInstitute's shape The quadrant of the coordinate system become keeps identical probability to exceed predetermined probabilities.

And, in V-type intersection point matching optimization module, described rejecting erroneous matching operates, and increases correct coupling, including following Unit,

First module, for initial threshold value d₁、d₂Initial value, current iteration frequency n=1；

Second unit, is used for deletingPresent in erroneous matching, anyIt is unsatisfactory for following two conditions will be deleted Remove:

j_qWith j '_qMeeting the constraint of core line, they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

j′_qWith j '_qHave and close on the distribution of V-type intersection topology and keep relative stability this characteristic；

Unit the 3rd, increases in the V-type intersection point not mated from two images and matchesIn, it is assumed that j and j ' is a pair point Not from the V-type intersection point not mated of two pictures, and if only if they when meeting following three condition, just think that (j, j ') is just Really coupling:

J and j ' meets the constraint of core line, and they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

The distance between feature description vector corresponding for j and j ' is less than certain predetermined threshold value d₂；

J and j ' has and closes on the distribution of V-type intersection topology and keep relative stability this characteristic；

Unit the 4th, for judging whether to meet the iteration termination condition preset, satisfied end iteration, order single line match Module works, and otherwise makes n=n+1, makes d₁=d₁+ c, order second unit works, and wherein, c is default step-length.

And, described iteration termination condition is, the matching number in current collection and last iteration perform after step 3.3 Join number to compare and do not increase, or current iteration number of times is more than preset times threshold value.

And, in single line match module, described based on local homography matrix estimate carry out matching treatment, it is achieved as follows,

Carry out local homography matrix to estimate, including the constraint utilizing the V-type intersection point of a pair coupling to be provided, obtain corresponding to this V The type intersection point homography matrix to the local matching region at place；

Mate single line segment, it is achieved as follows,

IfFor being mated institute's group by K the V-type intersection point obtained by above-mentioned V-type intersection point matching optimization module The set become,WithIt is illustrated respectively in the remaining line segment aggregate also not matched in two width figures；Right In each line segmentOrFind the V-type intersection point that closest 4 have matched, and this line segment is divided It is fitted in corresponding group；

Assuming that l and l ' is a pair line segment to be matched, be respectively allocated in the V-type intersection point j and j ' of coupling, wherein j is by line segment pair (l_m,l_n) formed, j ' by line segment to (l '_m,l′_n) formed, first determine whether the angle difference of l and l ' whether with line match (l_m,l′_m) and (l_n,l′_n) angle difference of middle conductor is similar；

If it is by the corresponding homography matrix H of the V-type intersection point (j, j ') of coupling, (l, l ') is evaluated, including by the two of l Intersection point is mapped in the plane of delineation at l ' place by H, obtains a new line segment l_hIf, l_hTwo end points to the distance of l ' WithSimultaneously less than respective threshold d_t, it is believed that (l, l ') is a pair correct line match.

The present invention converts at multiple image, suffers from good effect as under rotation, yardstick, visual angle change.Relative to existing Method, the technical program can be while finding more coupling, in most cases, keeps higher matching precision.

Accompanying drawing explanation

Fig. 1 is the flow chart of the embodiment of the present invention.

Fig. 2 is that the V-type intersection point of the embodiment of the present invention describes schematic diagram.

Fig. 3 is that the V-type intersection topology distribution relation that closes on of the embodiment of the present invention keeps stablizing schematic diagram before and after image converts.

Detailed description of the invention

Below by embodiment, technical scheme is described in further detail.

Assume that to be matched two image is expressed as I and I ', utilize existing line segments extraction method, from I and I ' middle extraction line segment, If the line segment aggregate extracted is respectivelyWithWherein M and N is illustrated respectively in the middle extraction of I and I ' The total number of line segment.The present invention is found by following steps and is present inWithIn line segment of the same name.

Step 1, V-type intersection point builds.

In each image, adjacent segments is intersected, form V-type intersection point.As a example by image I (in image I ' herewith Step), it is assumed that two lines sectionIf l_mAnd l_nFor adjacent segments, then they are intersected, form V-type intersection point, false It is set to j.After this step, V-type intersection point set will be respectively obtained I and I ' is middleWithWherein P and Q is illustrated respectively in two images the sum of the V-type intersection point formed.By couplingWithIn intersection point, can obtain Line match.If this is because intersection pointSuccessfully match intersection pointForm j_pTwo lines section will correspondingly with Form j '_qTwo lines section match.Therefore, for obtaining line match as much as possible, need to find as much as possible to be present inWithIn intersection point coupling.

Seeing Fig. 1, the line segments extraction result of image 1 obtains sets of line segments 1, by closing on line segment intersection, obtains V-type intersection point group 1, The line segments extraction result of image 2 obtains sets of line segments 2, by closing on line segment intersection, obtains V-type intersection point group 2.

Step 2, V-type intersection point mates.

First in setWithIn V-type intersection point carry out feature description, then by calculating the similarity between feature description Mate with some constraint antinodes.

Step 2.1, V-type intersection point describes.

As shown in Figure 2, close on line segment AB, CD and be crossed to form V-type intersection point O, take the Liang Ge concentric circular district centered by O Territory is characterized region, and wherein the radius of roundlet is the half of big radius of circle.The selection of radius of circle can be reasonable according to different application Choosing, bigger radius often makes to describe son and has higher resolving power, but the most sensitive to anamorphose.In an experiment Find, when the radius of roundlet elects 8 pixels as, there is preferable resultant effect.

The two concentric circular is divided into 4 regions by line segment OB and OD and their reverse extending line, and the most each region is again Being divided into a fan section and an annulus, this annular region is divided into the most again 3 parts.After this, the two is same The region at heart circle place is divided into 16 parts, and the area size at the most every 8 mutual diagonal angles is equal.8 are added up in these regions The pixel gradient rectangular histogram in individual direction, thus obtain 128 dimensional vectors description as this V-type intersection point.In order to make description Son has robustness to illumination variation, needs to be normalized the vector of gained.Yet with this 128 vector tieed up by two The rectangular histogram composition that group builds in the region of different area, thus when being normalized, need packet to be normalized, also Every 8 rectangular histograms i.e. built in area same area carry out normalization respectively.Meanwhile, in order to remove boundary effect, should In 128 dimensional vectors, the carrying out more than certain threshold value (such as 0.2) is blocked, to strengthen the effect of gradient direction.

Owing to the feature description vector of above V-type intersection point is to obtain by calculating grey level histogram in fixed size region, When there is dimensional variation when between image to be matched, the description vectors corresponding to same place can be very different, their feature Distance between description vectors thus can not correctly embody the similarity between them.The present invention proposes by building image pyramid Method solve this problem.

Image I and I ' is carried out the present invention Gaussian convolution and zoom operation obtains gaussian pyramid, while carrying out image conversion, The V-type intersection point built in original image is adjusted to accordingly in the image after change, and to the V after change in new image Type intersection point carries out the description of the characteristic area of fixed size as above.As a example by image I, it is being built gaussian pyramid During, when I becomes I_mTime,In V-type intersection point also do and be adapted to corresponding image change accordingly.False If when I becomes I_m, image down half, for certain V-type intersection pointAssume it by line segment to (l_x,l_y) constitute, it is suitable Picture size that should be new, line segment l_xAnd l_yLength also become original half, they constitute one new corresponding to j_pV-type hand over Point, it is assumed that forAt I_mIn, rightCarry out the feature description of fixed window size as above.In this manner, right In V-type intersection pointIts description will be indicated asWherein D_mRepresent j_pIt is being adjusted to image gold word Description vectors corresponding during tower m layer, M represents pyramidal total number of plies.

Step 2.2, V-type intersection point mates.

By previous step, V-type intersection point set can be obtainedWithIn the feature description of all V-type intersection points, for obtaining wherein Coupling, it is common practice to calculate the distance between they feature description vectors.But owing to V-type intersection point is by two adjacent segments Composition, the angle of cut of this two lines section keeps constant or small change under overwhelming majority image conversion.Thus V can be utilized This feature of type intersection point, weigh two from before the feature description similarity of the V-type intersection point of two pictures, by comparing shape The difference becoming their two lines section angle of cut tentatively judges whether they may be for correct couplings.If intersection point difference is corresponding more than preset Threshold value (when being embodied as, those skilled in the art can preset voluntarily, such as 30 °), being considered as the two V-type intersection point is not coupling. In this way, matching precision and the efficiency of algorithm are improved.

Owing to using by the way of building image pyramid, the V-type built in original image being handed over as described in step 2.1 Point is described, and the feature description of each V-type intersection point built in original graph will be made up of one group of vector, thus, evaluating During from the similarity of the feature description of two V-type intersection points in two width images, it should use special mode.

AssumeWithBeing a pair V-type intersection point to be matched from two width images, they are by based on image gold word After the description of tower, characteristic of correspondence describes and is respectivelyWithWherein D_mRepresent that j is corresponding to it Pyramidal m layer description son, D '_nRepresent the j ' description in its corresponding pyramidal n-th layer.By calculatingIn Each vector arrivesThe distance of middle institute directed quantity, obtainsThe feature description distance of j and j ' is defined asSeveral values of middle minimum (when being embodied as those skilled in the art can predetermined number voluntarily, such as 2) meansigma methods, this distance Function as the similarity weighed between j and j '.Carry out V-type intersection point description by building image pyramid and use characteristics described above Describing similarity evaluation mode and carry out V-type intersection point coupling, algorithm just can solve the chi being present between two images well Degree variation issue.This is because by building image pyramid, the V-type intersection point built in a width original image is passing through Gauss After convolution and zoom operation, a width can be adjusted to and be in the image of identical graphical rule with its V-type intersection point of the same name In, in this case, the similarity of the reaction the two V-type intersection point that the sub-distance of description between them can be real.

Step 3, V-type intersection point matching optimization.

V-type intersection point coupling obtained by step 2In be inevitably present erroneous matching, and wherein Miss out a part and there is correct coupling in both figures.This step is intended to optimizeIn coupling: reject erroneous matching same Time, increase correct coupling therein.In this process, the present invention proposes utilization and closes on the distribution of V-type intersection topology and keep relatively Stable method instructs erroneous matching delete and increase new coupling.

As shown in Figure 3, V-type intersection point j_cConstitute one with the two lines section and their extended line forming it and be similar to coordinate system Structure, with j_cThe most matched adjacent V-type intersection point (is j in this figure_1～8) be distributed in this coordinate system four quadrants (I, II, III, IV) in.j_cClose on V-type intersection point relative to j_cPosition image convert before and after should keep relative stability, this Present two aspects of liptinite: one, j_cNeighbours after converting through image are also its neighbours before image converts. After image converts, work as j_cBecome j '_c, with j '_cThe most matched V-type intersection point closed on will be at certain proportion (when being embodied as Those skilled in the art can preset ratio threshold value voluntarily, such as 0.5) on be j_cThe same place of neighbours.Reaction is exactly in fig. 3, At j '_1～8In these 8 points, wherein having a large portion is j_1～8The same place of these 8 points.Its two, for j_1～8With j '_1～8In Any pair of same place, wherein at j_cQuadrant in the coordinate system formed should be with another point at j ' on the biggest probability_cInstitute's shape The quadrant of the coordinate system become keeps identical.Such as, if j_mIt is j_1～8In a point, it is with j '_1～8In certain some j '_nIt is same place, Assume j_mIt is positioned at j_cThe first quartile of the coordinate system formed, then j '_nAlso should be in the biggest probability (this area skill when being embodied as Art personnel can predetermined probabilities threshold value voluntarily, such as 0.8) on be positioned at j '_cThe first quartile of the coordinate system formed.Above 2 is what is called Close on V-type intersection topology distribution keep relative stability.

Coupling obtained by utilization, can estimate the basis matrix between two width images with RANSAC, and the present invention proposes to pass through The mode of iteration is progressively rejectedIn erroneous matching and increase new coupling.Specifically comprise the following steps that

Step 3.1, initial threshold value d₁、d₂Initial value, current iteration frequency n=1；

Step 3.2, deletesPresent in erroneous matching.AnyIt is unsatisfactory for following two conditions will be deleted:

j_qWith j '_qMeeting the constraint of core line, i.e. they are relative to currentlyThe distance of corresponding basis matrix is less than certain predetermined threshold Value d₁；

j′_qWith j '_qHave and close on the distribution of V-type intersection topology and keep relative stability this characteristic.

Step 3.3, increases in the V-type intersection point not mated and match from two imageIn.Assume that j and j ' is to come respectively for a pair From the V-type intersection point not mated of two pictures, and if only if they when meeting following three condition, just think that (j, j ') is correct Join:

J and j ' meets the constraint of core line, and i.e. they are relative to currentlyThe distance of corresponding basis matrix is less than certain predetermined threshold Value d₁；

The distance between feature description vector corresponding for j and j ' is less than certain predetermined threshold value d₂；

J and j ' has and closes on the distribution of V-type intersection topology and keep relative stability this characteristic.

Note, in increasing coupling step, verify that a pair V-type intersection point coupling the most correctly mates many bars in step than deletion error Distance between two intersection point characteristic of correspondence description vectors of part is less than certain predetermined threshold value d₂.This is because at deletion error Join in step,In coupling to be evaluated certainly meeting this condition, be the most previously not increased toIn.

Step 3.4, it may be judged whether meet the iteration termination condition preset, be the matching number in current collection and upper in embodiment Matching number after secondary iteration execution step 3.3 is compared and is not increased, or current iteration number of times is more than preset times threshold value, example Such as 3, terminate iteration, enter step 4, otherwise make n=n+1, make d₁=d₁+ c, returns step 3.2 and continues executing with.Wherein, C is default step-length.

The dependent thresholds d related in above step₁、d₂, c all can be by this area practitioner rule of thumb sets itself.

Step 4, single line match.

The line segment carrying out mating not over formation V-type intersection point will mate in this step.Owing to these line segments are often from it His line segment is farther out, it is impossible to mates together with other line segments in the way of peering, thus is called single line segment in the present invention.

Step 4.1, locally homography matrix is estimated.

One homography matrix H has 8 degree of freedom, so that 8 independent constraints obtain unique solution.But, when two figures When basis matrix F between Xiang is known, there is following relation:

H^TF+F^TH=0

This equation provides 5 independent constraints, thus also needs to other 3 independent restrainings to obtain the unique solution of H.Present invention profit The constraint provided with a pair V-type coupling intersection point, obtains corresponding to this V-type intersection point the local matching region at place singly being answered square Battle array, method is as follows.

Assume that the locus of the camera corresponding to two images is expressed as C and C ', when C is expressed as [I | O], and wherein I is that order is The unit matrix of 3,0 is 0 corresponding vector；C ' can be expressed as [A | e '], and wherein e ' is the limit corresponding to the second pictures, A is remaining part, I, O, the decomposition of the corresponding camera matrix of A, e ' in addition to e ' after decomposing C '.Work as j_pWith j '_qCome In two width images a pair coupling intersection point, they respectively by line segment to (l_m,l_n) and (l '_m,l′_n) formed.In this case, by certain One space three-dimensional plane π=(v^T, 1) guided corresponding to j_pWith j '_qThe homography matrix H of the regional area at place can be expressed as:

H=A-e ' v^T

Because homography matrix H can set up point-to-point mapping relations, when assuming that x is l_mAn end points, after being mapped by H, it Same place x '=Hx will be at l '_mOn, that is:Accordingly, there exist following relation:

l′_m ^T(A-e′v^T) x=0

The equation is arranged, will obtain:

$x^{T}v=\frac{x^T{A}^T{l}^{\′}}{e^{\′T}{l}^{\′}}$

The equation is the linear equation of amount v to be asked, it is possible to ask v to provide a constraint, v for solution^TMatrix transpose for v.Due to two groups of line segments To (l_m,l_n) and (l '_m,l′_n) four these type of end points can be provided such that it is able to set up 4 equations as above, the least square solution of v Just can obtain from these 4 equations.After solving v, homography matrix H the most just can obtain accordingly.

Step 4.2, mates single line segment

AssumeK the V-type intersection point by being obtained by above-mentioned steps 3 mates the set formed,WithBeing illustrated respectively in the remaining line segment aggregate also not matched in two width figures, R, S are not the most not The line segment number mixed.For each line segmentOrThe present invention finds its closest 4 to match V-type intersection point, and this line segment is assigned in corresponding group.After this,WithIn line segment be assigned to different In group.In be assigned to the line segment in a group only and fromIts correspondence group in line segment mate, so can Reduce coupling interference, while improving precision, save the time.

Assuming that l and l ' is a pair line segment to be matched, they are respectively allocated in V-type coupling intersection point j and j ', and wherein j is by line segment pair (l_m,l_n) formed, j ' by line segment to (l '_m,l′_n) formed.First determine whether the angle difference of l and l ' whether with line match (l_m,l′_m) and (l_n,l′_n) angle difference of middle conductor is similar.Because l and l ' is to be assigned in the group at j and j ' place by the space relation of closing on, And line segment is after image converts adjacent to one another, their angle change is similar.According to such a fact, can be to (l, l ') Whether it is that correct coupling makes preliminary judgement.The differential seat angle assuming l and l ' is α, l_mWith l '_mDifferential seat angle be α_m, l_nWith l '_nAngle Difference is α_n, if there is:

$\left|\mathrm{\&alpha;}-\frac{a_m+{\mathrm{\&alpha;}}_n}{2}\right|<\mathrm{\&sigma;}$

Wherein σ is that angle threshold can be pre-set by those skilled in the art, as being set to 20 °.

If (l, l ') meets conditions above, the list estimated from V-type intersection point coupling (j, j ') followed by step 4.1 gained should Matrix, it is assumed that for H, (l, l ') is evaluated.Step 4.1 can also not performed when being embodied as, when needing, use step again The method of 4.1 estimates corresponding homography matrix.The two of l intersection points are mapped in the plane of delineation at l ' place by H, obtain one newly Line segment l_hIf (l, l ') is a pair correct line match, then l_hIt is necessarily located at the position that l ' closes on very much.Specifically, By calculating l_hTwo end points to the distance of l 'WithIfWithSimultaneously less than threshold value d_t, it is believed that (l, l ') meets this about Bundle.d_tCan be pre-set by those skilled in the art, as being set to 3 pixels.

Above determination methods can correctly judge whether a pair line match is correct coupling in most cases, but several when existing Bar line segment occur in same close region and the angle between them similar time, above method tends not to make accurate judgment, Invention increases a constraint to process problems.The brighter region of these two line segments of the same name of constraint requirements must be positioned at this Article two, the equidirectional of line segment.For example, if the left area of line segment l is brighter than its right area, that is the left side Area pixel gray average is greater than right area pixel grey scale average, then only in the left area of line segment l ' also than its right side When edge regions is wanted bright, just accepting (l, l ') is correct coupling.

The line match obtained by coupling single line and the line match obtained by coupling V-type intersection point together form final line segment Coupling.

When being embodied as, method provided by the present invention can realize automatic operational process based on software engineering, it is possible to uses modular mode Realize corresponding system.The present invention provides a kind of line match system estimated based on V-type intersection point and local homography matrix, passes through Arrange to find with lower module and be present inWithIn line segment of the same name,

V-type intersection point builds module, is respectively I and I ' for setting to be matched two image, extracts line segment from I and I ' respectively, if The line segment aggregate extracted is respectivelyWithWherein M and N is illustrated respectively in the middle extraction of I and I ' The total number of line segment, respectively obtains the set of V-type intersection point in image I and I 'WithWherein P and Q is illustrated respectively in two images the sum of the V-type intersection point formed；

V-type intersection point matching module, is used for including in setWithIn V-type intersection point carry out feature description, obtain V-type hand over The feature description vector of point, mates V-type intersection point by calculating the similarity between feature description vector, if coupling obtains The set that Q V-type intersection point is formed

V-type intersection point matching optimization module, for V-type intersection point matching module gainedUtilization closes on V-type The principle that intersection topology distribution keeps relative stability rejects erroneous matching, and increases correct coupling；

Single line match module, it is right to be used for includingWithIn not over formed V-type intersection point enter The line segment of row coupling, estimates to carry out matching treatment, gained line match result and V-type intersection point coupling based on local homography matrix excellent Change module gained line match result and constitute final line match result together.

And, in V-type intersection point matching optimization module, described in close on V-type intersection topology distribution keep relative stability, it is judged that mode It is as follows,

If V-type intersection point j_cA coordinate system is constituted, with j with the two lines section and line segment extended line forming it_cAdjacent is the most matched V-type intersection point be distributed in four quadrants of this coordinate system, when meeting following condition, it is judged that close on V-type intersection topology distribution Keep relative stability,

After image converts, work as j_cBecome j '_c, with j '_cThe most matched V-type intersection point closed on be j_cNeighbours same place surpass Cross preset ratio, and a pair same place of any of which, a little at j_cQuadrant in the coordinate system formed and another point are at j '_cInstitute's shape The quadrant of the coordinate system become keeps identical probability to exceed predetermined probabilities.

Further, in V-type intersection point matching optimization module, described rejecting erroneous matching operates, and increases correct coupling, including With lower unit,

First module, for initial threshold value d₁、d₂Initial value, current iteration frequency n=1；

Second unit, is used for deletingPresent in erroneous matching, anyIt is unsatisfactory for following two conditions will be deleted Remove:

j_qWith j '_qMeeting the constraint of core line, they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

j′_qWith j '_qHave and close on the distribution of V-type intersection topology and keep relative stability this characteristic；

Unit the 3rd, increases in the V-type intersection point not mated from two images and matchesIn, it is assumed that j and j ' is a pair point Not from the V-type intersection point not mated of two pictures, and if only if they when meeting following three condition, just think that (j, j ') is just Really coupling:

J and j ' meets the constraint of core line, and they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

The distance between feature description vector corresponding for j and j ' is less than certain predetermined threshold value d₂；

J and j ' has and closes on the distribution of V-type intersection topology and keep relative stability this characteristic；

Unit the 4th, for judging whether to meet the iteration termination condition preset, satisfied end iteration, order single line match Module works, and otherwise makes n=n+1, makes d₁=d₁+ c, order second unit works, and wherein, c is default step-length.

Each module implements and can be found in corresponding steps, and it will not go into details for the present invention.

It is emphasized that embodiment of the present invention is illustrative rather than determinate.Therefore the present invention includes also It is not limited to the embodiment described in detailed description of the invention, every is drawn according to technical scheme by those skilled in the art Other embodiments, also belong to the scope of protection of the invention.

Claims (10)

1. the line match method estimated based on V-type intersection point and local homography matrix, it is characterised in that: set to be matched two figure As respectively I and I ', from I and I ', extract line segment respectively, if the line segment aggregate extracted is respectivelyWith Wherein M and N is illustrated respectively in the total number of line segment of the middle extraction of I and I ', is found by following steps and is present inWithIn line segment of the same name,

Step 1, V-type intersection point builds, is included in the set respectively obtaining V-type intersection point in image I and I 'With The sum of the V-type intersection point formed during wherein P and Q is illustrated respectively in two images；

Step 2, V-type intersection point mates, including in setWithIn V-type intersection point carry out feature description, obtain V-type intersection point Feature description vector, by calculating the similarity between feature description vector, V-type intersection point is mated, if coupling obtains Q The set that individual V-type intersection point is formed

Step 3, V-type intersection point matching optimization, including to step 2 gainedUtilization closes on V-type intersection topology The principle that distribution keeps relative stability rejects erroneous matching, and increases correct coupling；

Step 4, single line match, including rightWithIn not over formed V-type intersection point carry out The line segment of coupling, estimates to carry out matching treatment, gained line match result and step 3 gained line segment based on local homography matrix Join result and constitute final line match result together.

The line match method estimated based on V-type intersection point and local homography matrix the most according to claim 1, it is characterised in that: step In rapid 3, described in close on V-type intersection topology distribution keep relative stability, it is judged that mode is as follows,

If V-type intersection point j_cA coordinate system is constituted, with j with the two lines section and line segment extended line forming it_cAdjacent is the most matched V-type intersection point be distributed in four quadrants of this coordinate system, when meeting following condition, it is judged that close on V-type intersection topology distribution Keep relative stability,

After image converts, work as j_cBecome j '_c, with j '_cThe most matched V-type intersection point closed on be j_cNeighbours same place surpass Cross preset ratio, and a pair same place of any of which, a little at j_cQuadrant in the coordinate system formed and another point are at j '_cInstitute's shape The quadrant of the coordinate system become keeps identical probability to exceed predetermined probabilities.

The line match method estimated based on V-type intersection point and local homography matrix the most according to claim 2, it is characterised in that: step In rapid 3, described rejecting erroneous matching, and increase correct coupling, it is achieved it is as follows,

Step 3.1, initial threshold value d₁、d₂Initial value, current iteration frequency n=1；

Step 3.2, deletesPresent in erroneous matching, anyIt is unsatisfactory for following two conditions will be deleted: j_qWith j '_qMeeting the constraint of core line, they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

j′_qWith j '_qHave and close on the distribution of V-type intersection topology and keep relative stability this characteristic；

Step 3.3, increases in the V-type intersection point not mated and match from two imageIn, it is assumed that j and j ' is to come respectively for a pair From the V-type intersection point not mated of two pictures, and if only if they when meeting following three condition, just think that (j, j ') is correct Join:

J and j ' meets the constraint of core line, and they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

The distance between feature description vector corresponding for j and j ' is less than certain predetermined threshold value d₂；

J and j ' has and closes on the distribution of V-type intersection topology and keep relative stability this characteristic；

Step 3.4, it may be judged whether meet the iteration termination condition preset, meets and terminates iteration, enters step 4, otherwise makes n=n+1, Make d₁=d₁+ c, returns step 3.2 and continues executing with, and wherein, c is default step-length.

The line match method estimated based on V-type intersection point and local homography matrix the most according to claim 3, it is characterised in that: institute Stating iteration termination condition is, the matching number after the matching number in current collection performs step 3.3 with last iteration is compared not There is increase, or current iteration number of times is more than preset times threshold value.

5. according to the line match method estimated based on V-type intersection point and local homography matrix described in claim 1 or 2 or 3 or 4, its Be characterised by: in step 4, described based on local homography matrix estimate carry out matching treatment, it is achieved as follows,

Step 4.1, carries out local homography matrix and estimates, including the constraint utilizing the V-type intersection point of a pair coupling to be provided, it is right to obtain Should be in this V-type intersection point homography matrix to the local matching region at place；

Step 4.2, mates single line segment, it is achieved as follows,

IfBy being mated, by K the V-type intersection point obtained by above-mentioned steps 3, the set formed,WithIt is illustrated respectively in the remaining line segment aggregate also not matched in two width figures；For each line SectionOrFind the V-type intersection point that closest 4 have matched, and this line segment is assigned to accordingly In group；

Assuming that l and l ' is a pair line segment to be matched, be respectively allocated in the V-type intersection point j and j ' of coupling, wherein j is by line segment pair (l_m,l_n) formed, j ' by line segment to (l '_m,l′_n) formed, first determine whether the angle difference of l and l ' whether with line match (l_m,l′_m) and (l_n,l′_n) angle difference of middle conductor is similar；

If it is by step 4.1 gained from the corresponding homography matrix H of the V-type intersection point (j, j ') of coupling, (l, l ') is evaluated, It is mapped in the plane of delineation at l ' place by H including by the two of l intersection points, obtains a new line segment l_hIf, l_hTwo ends Point is to the distance of l 'WithSimultaneously less than respective threshold d_t, it is believed that (l, l ') is a pair correct line match.

6. the line match system estimated based on V-type intersection point and local homography matrix, it is characterised in that: by arranging with lower module Find and be present inWithIn line segment of the same name,

V-type intersection point builds module, is respectively I and I ' for setting to be matched two image, extracts line segment from I and I ' respectively, if The line segment aggregate extracted is respectivelyWithWherein M and N is illustrated respectively in the middle extraction of I and I ' The total number of line segment, respectively obtains the set of V-type intersection point in image I and I 'WithWherein P and Q is illustrated respectively in two images the sum of the V-type intersection point formed；

V-type intersection point matching module, is used for including in setWithIn V-type intersection point carry out feature description, obtain V-type hand over The feature description vector of point, mates V-type intersection point by calculating the similarity between feature description vector, if coupling obtains The set that Q V-type intersection point is formed

V-type intersection point matching optimization module, for V-type intersection point matching module gainedUtilization closes on V-type The principle that intersection topology distribution keeps relative stability rejects erroneous matching, and increases correct coupling；

Single line match module, it is right to be used for includingWithIn not over formed V-type intersection point enter The line segment of row coupling, estimates to carry out matching treatment, gained line match result and V-type intersection point coupling based on local homography matrix excellent Change module gained line match result and constitute final line match result together.

The line match system estimated based on V-type intersection point and local homography matrix the most according to claim 6, it is characterised in that: V In type intersection point matching optimization module, described in close on V-type intersection topology distribution keep relative stability, it is judged that mode is as follows,

If V-type intersection point j_cA coordinate system is constituted, with j with the two lines section and line segment extended line forming it_cAdjacent is the most matched V-type intersection point be distributed in four quadrants of this coordinate system, when meeting following condition, it is judged that close on V-type intersection topology distribution Keep relative stability,

After image converts, work as j_cBecome j '_c, with j '_cThe most matched V-type intersection point closed on be j_cNeighbours same place surpass Cross preset ratio, and a pair same place of any of which, a little at j_cQuadrant in the coordinate system formed and another point are at j '_cInstitute's shape The quadrant of the coordinate system become keeps identical probability to exceed predetermined probabilities.

The line match system estimated based on V-type intersection point and local homography matrix the most according to claim 7, it is characterised in that: V In type intersection point matching optimization module, described rejecting erroneous matching operates, and increases correct coupling, including with lower unit,

First module, for initial threshold value d₁、d₂Initial value, current iteration frequency n=1；

Second unit, is used for deletingPresent in erroneous matching, anyIt is unsatisfactory for following two conditions will be deleted Remove:

j_qWith j '_qMeeting the constraint of core line, they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

j′_qWith j '_qHave and close on the distribution of V-type intersection topology and keep relative stability this characteristic；

Unit the 3rd, increases in the V-type intersection point not mated from two images and matchesIn, it is assumed that j and j ' is a pair point Not from the V-type intersection point not mated of two pictures, and if only if they when meeting following three condition, just think that (j, j ') is just Really coupling:

J and j ' meets the constraint of core line, and they are less than certain predetermined threshold value d relative to the distance of corresponding basis matrix₁；

The distance between feature description vector corresponding for j and j ' is less than certain predetermined threshold value d₂；

J and j ' has and closes on the distribution of V-type intersection topology and keep relative stability this characteristic；

Unit the 4th, for judging whether to meet the iteration termination condition preset, satisfied end iteration, order single line match Module works, and otherwise makes n=n+1, makes d₁=d₁+ c, order second unit works, and wherein, c is default step-length.

The line match system estimated based on V-type intersection point and local homography matrix the most according to claim 8, it is characterised in that: institute Stating iteration termination condition is, the matching number after the matching number in current collection performs step 3.3 with last iteration is compared not There is increase, or current iteration number of times is more than preset times threshold value.

10. according to the line match system estimated based on V-type intersection point and local homography matrix described in claim 6 or 7 or 8 or 9, its Be characterised by: in single line match module, described based on local homography matrix estimate carry out matching treatment, it is achieved as follows,

Carry out local homography matrix to estimate, including the constraint utilizing the V-type intersection point of a pair coupling to be provided, obtain corresponding to this V The type intersection point homography matrix to the local matching region at place；

Mate single line segment, it is achieved as follows,

IfFor being mated institute's group by K the V-type intersection point obtained by above-mentioned V-type intersection point matching optimization module The set become,WithIt is illustrated respectively in the remaining line segment aggregate also not matched in two width figures；Right In each line segmentOrFind the V-type intersection point that closest 4 have matched, and this line segment is divided It is fitted in corresponding group；

Assuming that l and l ' is a pair line segment to be matched, be respectively allocated in the V-type intersection point j and j ' of coupling, wherein j is by line segment pair (l_m,l_n) formed, j ' by line segment to (l '_m,l′_n) formed, first determine whether the angle difference of l and l ' whether with line match (l_m,l′_m) and (l_n,l′_n) angle difference of middle conductor is similar；

If it is by the corresponding homography matrix H of the V-type intersection point (j, j ') of coupling, (l, l ') is evaluated, including by the two of l Intersection point is mapped in the plane of delineation at l ' place by H, obtains a new line segment l_hIf, l_hTwo end points to the distance of l ' WithSimultaneously less than respective threshold d_t, it is believed that (l, l ') is a pair correct line match.