CN107622499A

CN107622499A - A kind of identification and space-location method based on target two-dimensional silhouette model

Info

Publication number: CN107622499A
Application number: CN201710734207.5A
Authority: CN
Inventors: 凌乐; 陈远强; 魏清平; 周东; 刘丝丝; 莫堃; 董娜; 于信宾
Original assignee: Dongfang Electric Corp
Current assignee: Dongfang Electric Corp
Priority date: 2017-08-24
Filing date: 2017-08-24
Publication date: 2018-01-23
Anticipated expiration: 2037-08-24
Also published as: CN107622499B

Abstract

The invention discloses a kind of identification based on target two-dimensional silhouette model and space-location method, it is related to technical field of image processing, for object be the object with two dimensional surface profile, identification target is the contour feature in some plane interested of the object, identification target is by a number of straightway and/or arc and/or the circular profile for connecting into closure, image can be from any angle, height photographic subjects and its residing background, but it must ensure that all profiles of target are visible and clear, then according to the two-dimensional mode profiles information of target, target is identified from image, and calculate locus of the target relative to camera.

Description

A kind of identification and space-location method based on target two-dimensional silhouette model

Technical field

The invention belongs to image processing field, is mainly used in the search from image and identifies the target of designated shape, and counts Spatial pose of the target relative to camera is calculated, the technology can be used for the scenes such as the orientation direction of robot, workpiece sorting.

Background technology

Template matches are the technologies for identifying and orienting that specified target is the most frequently used from image, that is, give the mould of a target Plate, then searched for and the most like object of template from image.Simplest template matching technique be from image find size, Direction and the completely the same object of template, it is only necessary to template is translated one by one along image pixel coordinates system, and calculated respectively similar Degree, until similarity is more than acceptance threshold, then it is assumed that search target；But the target and template in practical application in image Size is different and different due to shooting distance, while target might have rotation around face normal, and the above method is no longer applicable, But target does not deform in this case, i.e., circular or circular, rectangle is still rectangle, is only scaled by equal proportion , existing method is that pantograph ratio and the anglec of rotation are carried out into discretization, and original template is carried out into affine transformation according to certain step-length, Multiple different sizes, the template of different angle are generated, then carries out above-mentioned template-matching operation respectively again；Above-mentioned technology is in life In producing line to the automatic identification of workpiece and positioning can meet demand because height, the angle of camera and production line platform are Fixed, object is generally only had around the rotationally-varying of normal, however, when mechanical arm, robot need to carry out extraterrestrial target When identification and positioning, not only size, the normal anglec of rotation may change target in image, it is also possible to distortion of projection occurs, because It is arbitrary for shooting angle, now circle may be projected as ellipse, and rectangle is changed into parallelogram, by a plurality of straight line, camber line The irregular target of composition can then show the form being more difficult to, and above-mentioned template matching method is all no longer applicable.Throw Shadow deformation matching is the problem of image recognition, because it not only needs to identify target from the image of deformation, it is also necessary to calculate Go out the current locus of target (relative to camera), therefore the present invention is directed to distortion of projection's target, devises one kind and is based on mesh Mark identification and the space-location method of two-dimensional silhouette model.

The content of the invention

The object that the present invention is directed to is the object with two dimensional surface profile, and identification target is that some sense of the object is emerging Contour feature in interesting plane, identification target connect into closure by a number of straightway and/or arc and/or circle Profile, image can be from any angle, height photographic subjects and its residing background, but must ensure that all profiles of target are visible and clear It is clear, then according to the two-dimensional mode profiles information of target, target is identified from image, and calculate sky of the target relative to camera Between position.

The purpose of the present invention is achieved through the following technical solutions：

A kind of identification and space-location method based on target two-dimensional silhouette model, it is characterised in that comprise the following steps：

Step 1, prepare the two dimensional model of target, the true two dimensional model of target, long measure are drawn in graphics software For millimeter, and drawing is exported as to the file for including image outline information, two-dimensional mode profiles lines must close, profile by Straightway, arc or circle composition；

Step 2, the above-mentioned file for including image outline information is read, geometric primitive is parsed, searches and be depicted and close Profile is closed, and judges to distinguish ambiguity profile therein and non-ambiguity profile, the connection for describing closed contour cathetus and camber line is suitable Secondary relation, establish target two-dimensional silhouette geometric primitive topological relation；

Step 3, target image is gathered, using the edge of Canny edge detection operator detection images, travels through each wheel Exterior feature, non-closed profile is excluded, single-pass, more logical marginal point inspection are carried out to closed contour；Select single-pass, the more logical edge checked Point carries out intersecting outlines separation；Make polygonal segments closed contour, recurrence is carried out to each closed contour using Ramer algorithms Subdivision, until the ultimate range of obtained whole line segments to each self-corresponding closed contour is less than threshold value Dmax；Again to straightway And/or arc is fitted with merging, the friendship of adjacent segments is calculated according to the linear equation and/or elliptic arc equation fitted Point coordinates, the intersection point have corresponded to the target two-dimensional silhouette in step 2；

Step 4, profile is specified search for, caused all closed contours confirm and To Template profile in step 3 for retrieval The profile undetermined of identical line segment composition, treat the wide progress posture computing of fixed wheel and complete to confirm, marker recognition success in the picture Target, export target location and projection matrix.

In the step 2, the ambiguity profile is in the sequentially description of the profile, identical be present and sequentially describes；Establish The specific method of target two-dimensional silhouette geometric primitive topological relation is

Parse primitive information from the ENTITIES sections for the file for including image outline information, including Nl bar straight lines Start-stop coordinate, the center of circle of Na bar circular arcs, radius, start-stop coordinate, centre coordinate, the radius of Nc circle, preserve all straight lines, circular arc With round information；

Closed contour is searched, since the starting point of any straight line or circular arc, lookup overlaps down with the straight line terminal The starting point of one connecting line segment, and this process is repeated, until returning to the starting point most started, that is, find a closed contour；

The line segment included in the closed contour found is rejected from former data, repeats to search the step of closed contour Suddenly, finished until all line segments are searched, N number of closed contour is counted, plus the circle in objective contour being originally self-closing profile Nc, amounting in target two dimensional model has N+Nc closed contour, counts each closed contour cathetus number Nl_i, Circular Arc Na_i, circle number Nc_i, the numbering of wherein footmark i expression closed contours；

The connection sequentially relation of the straight line and camber line of non-circular closed contour in closed contour described, description rule is from appointing Meaning line segment starts, and returns to starting point along the fixed all line segments for traveling through profile clockwise or counterclockwise, is then designated as if straight line 0, camber line is designated as 1；For each non-circular closed contour, sequentially relation shares (NL for it_i+NA_i) kind, i.e., made with each line segment For starting point, obtain a kind of connection sequentially, complete the foundation of topological relation.

In the step 2, in addition to it is the checking step that ambiguity profile is also non-ambiguity profile to closed contour；Ambiguity wheel Exterior feature is in the sequentially description of the profile, identical be present and sequentially describes.

In the step 3, the method for excluding non-closed profile is：Travel through all marginal points, if some marginal point thirdly In neighborhood, the quantity of other marginal points is only 1, then the marginal point belongs to " patent chalaza ", after deleting the marginal point, reexamines The adjacent marginal point with " patent chalaza ", thirdly in neighborhood whether also only have other marginal points, can delete by that analogy Except a not closed profile；Said process is repeated, after having handled all marginal points, remaining profile is then all closed contour.

In the step 3, the method to closed contour single-pass, more logical marginal point inspections is the five neighborhood outermosts at a certain edge In making a circle, the quantity of non-adjacent marginal point is checked, being checked whether for the non-adjacent marginal point in everywhere can be along thirdly neighborhood Reach center edge point, if can if show at this centered on marginal point a passage, if can not if at this with center edge point It is unrelated, directly exclude, be otherwise single-pass for multi-channel edge point if the number of channels of the center edge point is more than 2 finally Road marginal point

In the step 3, intersecting outlines separation method is single-pass, the more logical marginal point to checking, from any marginal point Start, if the marginal point is single-pass marginal point, the next marginal point of an optional search channels, lead to marginal points if, then more Respectively along the next marginal point of each search channels；If having returned to the marginal point searched in search procedure, by the route Delete, return to starting point until searching for, more all paths for successfully returning to starting point, take most short one to be used as minimal closure wheel Exterior feature, said process is repeated, completes the separation to all intersecting outlines.

In the step 3, straightway and/or arc are fitted and to merge specific method as follows：

Oval segmental arc fitting, the adjacent both sides for the polygon for approaching closed contour are checked, using where ellipse fitting both sides Marginal point, calculate ellipse fitting error, if error is less than the approximate error of both sides and correspondence profile, replaced using elliptic arc Two straight lines, said process is repeated along adjacent two edges, until can not be preferably fitted using elliptic arc；

Oval segmental arc merges, if occurring in recursion is carried out to each closed contour using Ramer algorithms continuous Multistage uses the profile that elliptic arc is fitted, and calculates the center spacing of two adjacent ellipses arcs, if less than table if threshold value Dcmax Showing to merge, and after merging into an elliptic arc, repeat and carry out same operation with adjacent next elliptic arc, until phase Adjacent elliptic arc is disposed；

Fitting a straight line is ellipse to that can not be used in each closed contour progress recursion using Ramer algorithms with merging The profile that circular arc is fitted then directly is fitted using straight line, and carries out judgement merging to adjacent straightway；

The intersecting point coordinate of adjacent segments is sought, is AX according to linear equation Y=kX+b and elliptic arc equation²+BXY+CY²+DX+ EY+1=0 solves the intersecting point coordinate (X, Y) of adjacent segments, and the intersection point corresponds to the node of To Template；K, b are the Fitting Calculation in formula The coefficient gone out, only straight line is characterized, and (A, B, C, D, E) is the parameter that ellipse fitting calculates, each group of parameter characterization is only One is oval.

In the step 4, the method for specifying search for profile is, if multiple closed contours in objective contour be present, preferably Non- ambiguity profile is as template contours, if being all ambiguity profile, using the most profile of line segment number as template contours, from line segment The most profile of number all calculates similarity, selects most like one group along angle point.

In the step 4, the specific method for confirming the profile undetermined of line segment composition identical with To Template profile is retrieval Caused all closed contours in step 3, reject the composition line segment number profile different from template contours line segment；In remaining profile In, a node in the profile is arbitrarily selected, arranges the connection of the profile in the direction of the clock sequentially, if this sequentially can be Found in (NLi+NAi) of template contours in connection sequentially, then the search profile is profile undetermined, is otherwise directly excluded.

In the step 4, the method that profile undetermined carries out posture computing and completes to confirm is following steps：

According to profile undetermined and the sequentially relation of template contours, the corresponding pass of physics seat and pixel coordinate between node is determined System, template contours provide the physical coordinates of node, and the node of profile undetermined is pixel coordinate, initially sets up Attitude Calculation equation

WhereinFor the internal reference matrix of camera, can be obtained by camera calibration；

Optional three not conllinear nodes in template contours, by its physical coordinates [X_w Y_w 0]_i ^TAnd in profile undetermined The pixel coordinate [u0 v0] detected_i ^TBring above-mentioned Attitude Calculation equation into, obtain that projective transformation matrix is calculated

Projective transformation matrix is brought into above-mentioned Attitude Calculation equation, by all node physical coordinates [X of template contours_w Y_w 0]_i ^TBring above-mentioned Attitude Calculation equation into respectively, calculate the projected pixel coordinate [u v] of each node_i ^T；

Projection similarity F is calculated, according to the corresponding node detected in node pixel coordinate (u v) i and profile undetermined Pixel coordinate (u0v0) i calculates similarity F, and calculation formula is as follows

Wherein n represents nodal point number summation in template contours；

For non-ambiguity profile, if similarity exceedes given threshold, then it is assumed that the profile undetermined is correct target, and is returned Return to projective transformation squareBattle array, the matrix then describe spatial translation position of the target with respect to camera and Its anglec of rotation around each axle；

For ambiguity profile, if the similarity under a variety of node corresponding relations exceedes given threshold, will obtain multiple Attitude matrix, then the profile undetermined is set to polymorphic profile, if target only has this ambiguity profile, directly returns and conform to The multiple attitude matrixs asked；

If the common attitude matrix of existence anduniquess between multiple polymorphic profiles, the attitude matrix is the correct appearance of ambiguity profile State；If still there is multiple identical attitude matrixs, the distribution that illustrates multiple ambiguity profiles of target be it is full symmetric, then directly Connect and return to multiple attitude matrixs, and the successful target of marker recognition in the picture.

Beneficial effects of the present invention are as follows：

First, a kind of identification and space-location method based on target two-dimensional silhouette model provided by the invention, the two of target Although profile target is tieed up after distortion of projection, the length, radius, angle, proportionate relationship etc. of geometric primitive (straightway, arc) Change is may have occurred, but keeps the topological relation between invariably geometric primitive；Topological relation is the objective contour bag The number of straight segments that contains, camber line segment number, circle quantity, the connection sequentially relation of straightway and arc；According to setting the goal Skeleton pattern, establishes the topological relation of its geometric primitive, then treats search image and carries out rim detection, then carries out geometric primitive Segmentation fitting, by all contours segmentations be straight line, ellipse, elliptic arc, according to the topological relation of object module, searched into image Rope matches, untill the topological relation of target is met.

2nd, a kind of identification and space-location method based on target two-dimensional silhouette model provided by the invention, use Canny is the generally acknowledged best pixel edge detection algorithm of industry, and Ramer algorithms are similarly.

3rd, a kind of identification and space-location method based on target two-dimensional silhouette model provided by the invention, target is identified Contour feature as in some plane interested of the object.Identify target by a number of straightway and (or) arc And (or) the circular profile for connecting into closure.Image can be from any angle, height photographic subjects and its residing background, but must ensure All profiles of target are visible and clear, then according to the two-dimensional mode profiles information of target, target identified from image, and count Calculate locus of the target relative to camera.

Brief description of the drawings

Fig. 1 is that target image Internal periphery of the present invention describes schematic diagram；

Fig. 2 is the schematic diagram in three fields of the invention；

Fig. 3 is the schematic diagram in five fields of the invention；

Fig. 4 is intersecting outlines Shortest Path Searching of the present invention with separating tactful schematic diagram；

Fig. 5 is the logical construction schematic diagram of the present invention.

Embodiment

Further illustrated below by way of several specific embodiments to realize the technical scheme of the object of the invention, it is necessary to illustrate , claimed technical solution of the invention includes but is not limited to following examples.

Embodiment 1

Such as Fig. 1 to 5, a kind of identification and space-location method based on target two-dimensional silhouette model, it is characterised in that including Following steps：

This is a kind of most basic embodiment of the present invention.Although the two-dimensional silhouette target of target is after distortion of projection, several The length, radius, angle, proportionate relationship etc. of what primitive (straightway, arc) may have occurred change, but keep constant It is the topological relation between geometric primitive；Topological relation be the objective contour include number of straight segments, camber line segment number, circle number Amount, the connection sequentially relation of straightway and arc；According to the skeleton pattern to set the goal, the topology for establishing its geometric primitive is closed System, then treat search image and carry out rim detection, then carry out the segmentation fitting of geometric primitive, be straight by all contours segmentations Line, ellipse, elliptic arc, according to the topological relation of object module, matching is searched for into image, until meeting the topological relation of target Untill；The Canny of use is the generally acknowledged best pixel edge detection algorithm of industry, and Ramer algorithms are similarly.

Embodiment 2

A kind of identification and space-location method based on target two-dimensional silhouette model, it is characterised in that comprise the following steps

Step 1, prepare the two dimensional model of target, the true two dimensional model of target is drawn in AutoCAD, long measure is Millimeter, and DXF formatted files are exported as, two-dimensional mode profiles lines must close, and profile is by straightway, arc or circle group Into；

Step 2, the file for including image outline information is read, geometric primitive is parsed, it is several to establish target two-dimensional silhouette What primitive topological relation, searches and closed contour is depicted, and determine whether ambiguity profile；

Step 3, target image is gathered, using edge in Canny edge detection operator detection images, excludes non-closed wheel Exterior feature, single-pass, more logical marginal point inspection are carried out to closed contour, and the single-pass for selecting to check, more logical marginal point carry out intersecting outlines Separation, recursion is carried out to each closed contour using Ramer algorithms, until obtained whole line segments are to each self-corresponding Untill the ultimate range of contour segment is less than a certain threshold value Dmax；Straightway and/or arc are fitted with merging again, according to The linear equation and/or elliptic arc equation fitted calculates the intersecting point coordinate of adjacent segments, and the intersection point has corresponded to To Template Node；

Closed contour is searched, since the starting point of any straight line or circular arc, lookup overlaps down with the straight line terminal The starting point of one connecting line segment, and this process is repeated, until returning to the starting point most started, that is, find a closed contour； Because intersecting situation is not present in the closed contour of To Template, therefore the searching route only one of starting point can be returned to.Special instruction, Below, when closed contour is searched from real image, it is understood that there may be intersecting closed contour, now need to search by shortest path Rope closed contour, refer to aftermentioned.

The line segment included in the closed contour found is rejected from former data, repeats to search the step of closed contour Suddenly, finished until all line segments are searched, count N number of closed contour, be originally self-closing wheel plus the circular Nc in objective contour Exterior feature, therefore have (N+Nc) individual closed contour altogether in target two dimensional model, count each closed contour cathetus number NLi, circle Arc number NAi, circle number NCi, wherein footmark i represent the numbering of closed contour；

Sequentially relation, description rule are since any line segment, along admittedly for the connection of description closed contour cathetus and camber line The fixed all line segments for traveling through profile clockwise or counterclockwise return to starting point, are then designated as 0 if straight line, camber line is designated as 1；For Non-circular profile, sequentially relation shares (NLi+NAi) kind for it, i.e., using each line segment as starting point, obtains a kind of connection sequentially. Such as in Fig. 1 examples, the connection of outline is sequentially represented by 1-0-0-0,0-1-0-0,0-0-1-0, tetra- kinds of 0-0-0-1.

Check whether as " ambiguity profile "；" ambiguity profile " is during (NLi+NAi) kind of the profile sequentially describes, phase be present Same sequentially description, in Fig. 1 examples, four kinds of connections of Internal periphery are sequentially 1-0-1-0,0-1-0-1,1-0-1-0,0-1-0- 1；" ambiguity profile " can not obtain unique match result in the matching process.

Searched for since the closed contour of " non-ambiguity ", if being all " having ambiguity " closed contour, most from angle point number Closed contour starts along corners Matching, shares X kind corresponding relations, all calculates similarity, select most like one group；

In the step 3, the method for non-closed profile is excluded to travel through all marginal points, if some marginal point is thirdly adjacent In domain, the quantity of other marginal points is only 1, then the marginal point belongs to " patent chalaza ", after deleting the marginal point, reexamine with " patent chalaza " adjacent marginal point, thirdly in neighborhood whether also only have other marginal points, can delete by that analogy One not closed profile；Said process is repeated, after having handled all marginal points, remaining profile is then all closed contour.

In the step 3, the method to closed contour single-pass, more logical marginal point inspections is the five neighborhood outermosts at a certain edge In making a circle, the quantity of non-adjacent marginal point is checked, being checked whether for the non-adjacent marginal point in everywhere can be along thirdly neighborhood Reach center edge point, if can if show at this centered on marginal point a passage, if can not if at this with center edge point It is unrelated, directly exclude, be otherwise single channel side for multi-channel edge point if the passage of the center edge point is more than 2 finally Edge point

In the step 3, intersecting outlines separation method is single-pass, the more logical marginal point to checking, from any marginal point Start, if the marginal point is single-pass, the next marginal point of an optional search channels, if multi-channel edge point, then distinguish Along the next marginal point of each search channels；If having returned to the marginal point searched in search procedure, the route is deleted, Until search returns to starting point, more all successfully return to expects your path, takes most short one to be used as minimal closure profile, weight Multiple said process, completes the separation to all intersecting outlines.

In the step 3, straightway and/or arc are fitted and to merge specific method as follows

Oval segmental arc fitting, the adjacent both sides of Close polygon are checked, use the marginal point where ellipse fitting both sides, meter Ellipse fitting error is calculated, if error is less than the approximate error of both sides and correspondence profile, two straight lines are replaced using elliptic arc. Said process is repeated along adjacent two edges, until can not be preferably fitted using elliptic arc.

Oval segmental arc merges, if occurring the profile that continuous multi-stage uses elliptic arc to be fitted in step f, consideration is It is no to can be merged into an elliptic arc to be fitted.Specific method is：The center spacing of two adjacent ellipses arcs is calculated, if being less than threshold Value Dcmax then represents to merge, and after merging into an elliptic arc, repeats and is carried out equally with adjacent next elliptic arc Operation, until adjacent ellipses arc is disposed.

With merging, the profile that elliptic arc can not be used to be fitted then directly is fitted fitting a straight line using straight line, and Judgement merging is carried out to adjacent straightway.

The intersecting point coordinate of adjacent segments is sought, adjacent segments are calculated according to the above-mentioned linear equation fitted, elliptic arc equation Intersecting point coordinate, the intersection point corresponded to the node of To Template.

In the step 4, the method for specifying search for profile is, if multiple closed contours in objective contour be present, preferably " non-ambiguity " profile is as template contours, if " non-ambiguity " profile is not present, using the most profile of line segment number as template wheel It is wide.

In the step 4, the specific method for confirming the profile undetermined of line segment composition identical with To Template profile is retrieval Caused all closed contours in step 3, reject the composition line segment number profile different from template contours line segment；In remaining profile In, the node (intersection points of adjacent segments) in the profile is arbitrarily selected, by (consistent with the direction in step 2e) clockwise Direction arranges the connection of the profile sequentially, if this can sequentially find in (NLi+NAi) of template contours in connection sequentially, The search profile is profile undetermined, is otherwise directly excluded.

In the step 4, the method that profile undetermined carries out posture computing and completes to confirm is following steps

According to profile undetermined and the sequentially relation of template contours, the corresponding relation of " physics-pixel " between node, mould are determined Web wheel exterior feature provides the physical coordinates of node, and the node of profile undetermined is pixel coordinate, has if template contours are " ambiguity profile " It two or more corresponding relations met, should put on an equal footing, complete following calculating

Initially set up Attitude Calculation equation：

Wherein n represents nodal point number summation in template contours；

For " non-ambiguity profile ", if similarity exceedes given threshold, then it is assumed that the profile undetermined is correct target, and Return to projective transformation matrixThe matrix then describes spatial translation position of the target with respect to camera And its anglec of rotation around each axle；

For " ambiguity profile ", if the similarity under a variety of node corresponding relations exceedes given threshold, will obtain multiple Attitude matrix, can not uniquely determine, then the profile undetermined is set to " polymorphic profile ", if target only has this " ambiguity wheel It is wide ", then directly return to satisfactory multiple attitude matrixs；

If target also has other " ambiguity profiles ", other profiles are set as new template contours, at " polymorphic profile " Nearby repeat the above steps to obtain new " polymorphic profile "；

If the common attitude matrix of existence anduniquess between multiple " polymorphic profiles ", the attitude matrix be " ambiguity profile " just True posture；If still there is multiple identical attitude matrixs, the distribution for illustrating multiple " the ambiguity profiles " of target is full symmetric , then directly return to multiple attitude matrixs, and the successful target of marker recognition in the picture.

Embodiment 3

A kind of identification and space-location method based on target two-dimensional silhouette model, comprise the following steps

Step 1, prepare the two dimensional model of target, the true two dimensional model of target, long measure are drawn in graphics software For millimeter, and drawing is exported as into DXF formatted files, two-dimensional mode profiles lines must close, and profile is by straightway, arc Or circle composition；

Step 3, target image is gathered, using the edge of Canny edge detection operator detection images, travels through each wheel Exterior feature, non-closed profile is excluded, single-pass, more logical marginal point inspection are carried out to closed contour, the edge of image is the pixel on edge Point forms a series of profile, and marginal point refers in step 3 with the pixel on the edge of Canny edge detection operator detection images Point, that is, the pixel on profile；Single-pass, the more logical marginal point for selecting to check carry out intersecting outlines separation；Make polygon Closed contour is approached, recursion is carried out to each closed contour using Ramer algorithms, until obtained whole line segments are to respectively The ultimate range of self-corresponding closed contour is less than threshold value Dmax；Straightway and/or arc are fitted with merging again, root The intersecting point coordinate of adjacent segments is calculated according to the linear equation and/or elliptic arc equation fitted, the intersection point has been corresponded in step 2 Target two-dimensional silhouette, and be made up of geometric primitives such as straightway and arcs, node is exactly target two-dimensional silhouette cathetus The intersection point of section and/or oval arc；Canny is the generally acknowledged best pixel edge detection algorithm of industry, and Ramer algorithms are same Reason；Threshold value Dmax suggests taking 2, suggested according to for Ramer algorithm authors in paper；" single-pass, more logical marginal point " is used to describe Any edge pixel relation adjacent with periphery edge point, for judging whether profile intersects；

The line segment included in the closed contour found is rejected from former data, repeats to search the step of closed contour Suddenly, finished until all line segments are searched, N number of closed contour is counted, plus the circle in objective contour being originally self-closing profile Nc, amounting in target two dimensional model has N+Nc closed contour, counts each closed contour cathetus number Nli, Circular Arc Nai, circle number Nci, wherein footmark i represent the numbering of closed contour；

The connection sequentially relation of the straight line and camber line of non-circular closed contour in closed contour described, description rule is from appointing Meaning line segment starts, and returns to starting point along the fixed all line segments for traveling through profile clockwise or counterclockwise, is then designated as if straight line 0, camber line is designated as 1；For each non-circular closed contour, sequentially relation shares (NLi+NAi) kind for it, i.e., is made with each line segment For starting point, obtain a kind of connection sequentially, complete the foundation of topological relation.

In the step 2, in addition to it is the checking step that ambiguity profile is also non-ambiguity profile to closed contour；Ambiguity wheel Exterior feature is in the sequentially description of the profile, identical be present and sequentially describes, in Fig. 1 examples, four kinds of connections of Internal periphery are sequentially 1-0-1-0,0-1-0-1,1-0-1-0,0-1-0-1；Unique corresponding relation is not present in ambiguity profile, therefore can not subsequently obtain Unique solution adds amount of calculation, therefore non-ambiguity is preferentially used if having non-ambiguity profile, it is necessary to further excluded Profile carries out matching primitives, if being all ambiguity profile, carries out matching primitives according still further to another method, the calculating that both need Journey is differentiated, it is necessary to make a distinction；

Three neighborhoods refer to centered on marginal point to be calculated, and 3 multiply 3 pixel window, such as Fig. 2, and similarly " five neighborhoods " refers to treat Centered on calculating marginal point, 5 multiply 5 pixel window, such as Fig. 3；" thirdly in neighborhood, the quantity of other marginal points is only 1 " to refer to One and only one pixel is the side for the current outline for belonging to traversal in eight in addition to itself pixel in three neighborhoods Edge point, as shown in Fig. 2 black represents marginal point, white is the pixel of non-edge point；Such as Fig. 3, " non-adjacent marginal point " refers to During five neighborhood outermosts make a circle, the pixel of the 3X3 inside five neighborhoods is not considered now, if it is adjacent in pixel that outermost is made a circle A dry pixel regards a marginal point as, but if inherently isolated in the presence of some pixel, not with any other picture Vegetarian refreshments is adjacent, then is directly denoted as a marginal point.The marginal point for being referred to as above-mentioned two situations is " non-adjacent marginal point ", such as under In figure left figure, labeled as 1,2 marginal point, inherently isolated in outermost makes a circle, therefore be designated as " non-neighboring edge fit respectively Edge point ", and labeled as 3 place, it is adjacent to have two marginal points, therefore is regarded as " marginal point ", such left figure In a total of three " non-adjacent marginal point "；" passage of center edge point " can be along phase if referring to above-mentioned " non-adjacent marginal point " Adjacent side edge point search communicates to center edge point, i.e. " non-adjacent marginal point " with center edge point, is designated as a passage.

In the step 3, intersecting outlines separation method is single-pass, the more logical marginal point to checking, from any marginal point Start, if the marginal point is single-pass marginal point, the next marginal point of an optional search channels, lead to marginal points if, then more Respectively along the next marginal point of each search channels；If having returned to the marginal point searched in search procedure, by the route Delete, until search returns to starting point, more all paths for successfully returning to starting point, path is closed for that can be formed along edge point search The profile of conjunction, take most short one to be used as minimal closure profile, repeat said process, complete the separation to all intersecting outlines.

Here threshold value Dcmax based on experience value can be using the scope of song as 0.5~2.

According to profile undetermined and the sequentially relation of template contours, the corresponding relation of " physics-pixel " between node, mould are determined Web wheel exterior feature provides the physical coordinates of node, and the node of profile undetermined is pixel coordinate, has two if template contours are ambiguity profile The corresponding relation that kind or more kind meets, should put on an equal footing, complete following calculating

Initially set up Attitude Calculation equation

Wherein n represents nodal point number summation in template contours；

Given threshold describes the mean pixel allowable deviation for calculating node and theoretical node, it is proposed that 1 to 5 is taken, if matching It is required that strict, then threshold value should be smaller, otherwise bigger.

If target also has other ambiguity profiles, other profiles are set as new template contours, near polymorphic profile Repeat the above steps to obtain new polymorphic profile；

For the first time polymorphic Contour searching being carried out to scan for from image overall, amount of calculation is relatively large, therefore rear When continuing other polymorphic Contour searchings, in order to reduce amount of calculation, can near the position of a upper polymorphic profile search for.It is more State profile nearby i.e. more than the center of a polymorphic profile be the center of circle, gradually increase search radius, outside diffused searching other can The closed contour of energy, the parameter searched in calculating process are identical with the last time.

Embodiment 4

Such as Fig. 1 to Fig. 5, a kind of identification and space-location method based on target two-dimensional silhouette model, it is characterised in that bag Include following steps：

1. prepare the two dimensional model of target.The true two dimensional model of target is drawn in AutoCAD, long measure is milli Rice, and export as DXF formatted files, it is desirable to objective contour must be closed, and profile is made up of straightway, arc or circle.Such as figure Shown in example 1, the target has four closed contours, is made up of straightway, circle and circular arc.

2. reading DXF file, parsing geometric primitive, and establish target two-dimensional silhouette geometric primitive topological relation.

A) primitive information is parsed from the ENTITIES sections of DXF files, the start-stop mainly including straight line (note Nl bars) is sat Mark, the center of circle of circular arc (note Na bars), radius, start-stop coordinate, centre coordinate, the radius of circle (note Nc), preserve all straight lines, circle Arc, circle information.

B) closed contour is searched.Since the starting point of any straight line (or circular arc), lookup overlaps with the straight line terminal Next connecting line segment starting point, repeat a upper process, up to returning to the starting point most started, i.e., successfully find a closure Profile.(because intersecting situation is not present in the closed contour of To Template, therefore the searching route only one of starting point can be returned to.It is special Do not mentionlet alone it is bright, below, when closed contour is searched from real image, it is understood that there may be intersecting closed contour, now need by most Short-circuit path search closed contour, after referring to.)

C) line segment included in the closed contour found is rejected from former data, repeat step b, until will be all Line segment is searched and finished, and counts N number of closed contour.Originally it is self-closing profile plus the circular Nc in objective contour, therefore target two Amounting in dimension module has (N+Nc) individual closed contour.

D) each closed contour cathetus number NLi, Circular Arc NAi, circle number NCi, wherein footmark i expression closure wheel are counted Wide numbering.

E) connection of closed contour cathetus and camber line sequentially relation is described.Description rule is the edge since any line segment All line segments of fixed (or counterclockwise) clockwise direction traversal profile return to starting point, are then designated as 0 if straight line, camber line is designated as 1. For non-circular profile, sequentially relation shares (NLi+NAi) kind for it, i.e., using each line segment as starting point, obtains a kind of connection Sequentially.Such as in Fig. 1 examples, the connection of outline is sequentially represented by 1-0-0-0,0-1-0-0,0-0-1-0,0-0-0-1 tetra- Kind.

F) check whether as " ambiguity profile "." ambiguity profile " is during (NLi+NAi) kind of the profile sequentially describes, to exist Identical sequentially describes, and in Fig. 1 examples, four kinds of connections of Internal periphery are sequentially 1-0-1-0,0-1-0-1,1-0-1-0,0-1- 0-1." ambiguity profile " can not obtain unique match result in matching process later.

3. IMAQ and processing, establish geometric topo-relationship.

A) target image is gathered, feature to be identified must be completely clear, and the actual physical size of feature must meet it as far as possible The design size of two dimensional model.

B) edge in Canny edge detection operator detection images is used, according to image actual conditions, suitably reduces strong edge Threshold value, the object edge for ensureing be detected occur without missing inspection, make continuous edge as far as possible, extra increased short weak edge Excluded in next step.

C) non-closed profile is excluded.The method for excluding non-closed profile is：

All marginal points are traveled through, if some marginal point, thirdly in neighborhood, the quantity of other marginal points is only 1, then the side Edge point belongs to " patent chalaza ", after deleting the marginal point, reexamine with " patent chalaza " adjacent marginal point, thirdly neighborhood In whether also only have other marginal points, a not closed profile can be deleted by that analogy.Repeat said process, processing After complete all marginal points, remaining profile is then all closed contour.

D) single-pass, more logical marginal point inspection.After step c exclusion, remaining profile is closed contour in image, From in terms of neighbouring relations, can be divided between profile it is intersecting, mutually from.For phase from closed contour, the node of single-pass in profile be present, and For intersecting profile, wherein more logical marginal points (node) be present, the determination methods of single-pass marginal point and Duo Tong marginal points：At certain During the five neighborhood outermosts at one edge make a circle, check that non-adjacent marginal point there are several places, for everywhere, whether reexamine can edge Thirdly neighborhood reaches center edge point, if can if show at this centered on marginal point a passage, if can not if at this with Heart marginal point is unrelated, directly excludes.Finally, if the passage of the center edge point is more than 2, for multi-channel edge point, otherwise for Single channel marginal point, as shown in Figure 2.

E) intersecting outlines separate.According to the single-pass marginal point (node) judged in step d and more logical marginal point (node), Since any marginal point, if the marginal point is single-pass, the next marginal point of an optional search channels, if multichannel side Edge point, then respectively along the next marginal point of each search channels；, will if having returned to the marginal point searched in search procedure The route is deleted, and until search returns to starting point, more all successfully return to expects your path, takes most short one as minimum Closed contour, said process is repeated, the separation to all intersecting outlines is completed, as Fig. 3 shows.

F) polygonal segments closed contour.Recursion is carried out to each closed contour using Ramer algorithms, until To whole line segments to each self-corresponding contour segment ultimate range be less than a certain threshold value Dmax untill.

G) oval segmental arc fitting.The adjacent both sides of Close polygon are checked, using the marginal point where ellipse fitting both sides, Ellipse fitting error is calculated, if error is less than the approximate error of both sides and correspondence profile, two straight lines are replaced using elliptic arc Side.Said process is repeated along adjacent two edges, until can not be preferably fitted using elliptic arc.

H) oval segmental arc merges.If occurring the profile that continuous multi-stage uses elliptic arc to be fitted in step f, consider Whether an elliptic arc can be merged into be fitted.Specific method is：The center spacing of two adjacent ellipses arcs is calculated, if being less than Threshold value Dcmax then represents to merge, and after merging into an elliptic arc, repeats and is carried out equally with adjacent next elliptic arc Operation, until adjacent ellipses arc is disposed.

I) fitting a straight line is with merging.In step f, the profile that elliptic arc can not be used to be fitted then directly uses straight line It is fitted, specific approximating method is not in scope.And judgement merging is carried out to adjacent straightway.

J) intersecting point coordinate of adjacent segments is sought.Adjacent lines are calculated according to the above-mentioned linear equation fitted, elliptic arc equation The intersecting point coordinate of section, the intersection point have corresponded to the node of To Template.

4. searched for since the closed contour of " non-ambiguity ".It is most from angle point number if being all " having ambiguity " closed contour Closed contour start along corners Matching, share X kind corresponding relations, all calculate similarities, select most like one group.

A) profile is specified search for.If multiple closed contours in objective contour be present, preferred " non-ambiguity " profile is as mould Web wheel is wide, if " non-ambiguity " profile is not present, using the most profile of line segment number as template contours.

B) caused all closed contours, rejecting composition line segment number are different from template contours line segment in step 3 for retrieval Profile.In remaining profile, some node (intersection points of adjacent segments) in the profile is arbitrarily selected, by clockwise (with step Direction in rapid 2e is consistent) direction arranges the connection of the profile sequentially, if this sequentially can be in (NLi+NAi) of template contours Found in connection sequentially, then the search profile is profile undetermined, is otherwise directly excluded.Profile undetermined only possesses at present and template Profile identical line segment forms, but might not comply fully with the true shape of target, therefore also needs to carry out Similarity Measure.

C) according to profile undetermined and the sequentially relation of template contours, the corresponding relation of " physics-pixel " between node is determined, Template contours provide the physical coordinates of node, and the node of profile undetermined is pixel coordinate.If template contours are " ambiguity profile " There is the corresponding relation that two or more meet, should put on an equal footing, complete following calculating.Initially set up following Attitude Calculation side Journey：

WhereinFor the internal reference matrix of camera, can be obtained by camera calibration.Optional three in template contours Not conllinear node, by its physical coordinates [X_w Y_w 0]_i ^TAnd the pixel coordinate [u0 v0] detected in profile undetermined_i ^TBring into Formula 2, obtain that projective transformation matrix is calculated

D) projective transformation matrix obtained in step c is brought into formula 1, by all node physical coordinates [X of template contours_w Y_w 0]_i ^TBring formula 1 into respectively, calculate the projected pixel coordinate [u v] of each node_i ^T。

E) projection similarity F is calculated.According to node pixel coordinate (u v) i calculated in step d with being examined in profile undetermined Pixel coordinate (u0 v0) i for the corresponding node measured calculates similarity F, and calculation formula is as shown in Equation 3.

Wherein n represents nodal point number summation in template contours

F) for " non-ambiguity profile ", if similarity exceedes given threshold, then it is assumed that the profile undetermined is correct target, And return to its projective transformation matrix being calculated in step cThe matrix then describes target phase Spatial translation position to camera and its anglec of rotation around each axle.

G) for " ambiguity profile ", if the similarity under a variety of node corresponding relations exceedes given threshold, will obtain more It individual attitude matrix, can not uniquely determine, then be set to the profile undetermined " polymorphic profile ".If target only has this " ambiguity Profile ", then directly return to satisfactory multiple attitude matrixs.

If h) target also has other " ambiguity profiles ", other profiles are set as new template contours, in " polymorphic wheel It is wide " neighbouring repeat step b~e.New " polymorphic profile " will be obtained.

If the common attitude matrix of existence anduniquess between i) multiple " polymorphic profiles ", the attitude matrix is " ambiguity profile " Correct posture；If still there is multiple identical attitude matrixs, the distribution for illustrating multiple " the ambiguity profiles " of target is completely right Claim, then directly return to multiple attitude matrixs, and the successful target of marker recognition in the picture.

Claims

1. a kind of identification and space-location method based on target two-dimensional silhouette model, it is characterised in that comprise the following steps：

Step 1, prepare the two dimensional model of target, the true two dimensional model of target is drawn in graphics software, long measure is milli Rice, and drawing is exported as to the file for including image outline information, two-dimensional mode profiles lines must close, and profile is by straight line Section, arc or circle composition；

Step 2, the above-mentioned file for including image outline information is read, parses geometric primitive, searches and closure wheel is depicted Exterior feature, and judge to distinguish ambiguity profile therein and non-ambiguity profile, the connection for describing closed contour cathetus and camber line is sequentially closed System, establishes target two-dimensional silhouette geometric primitive topological relation；

Step 3, target image is gathered, using the edge of Canny edge detection operator detection images, travels through each profile, row Unless closed contour, single-pass, more logical marginal point inspection are carried out to closed contour；Single-pass, the more logical edge for selecting to check click through Row intersecting outlines separate；Make polygonal segments closed contour, it is thin to carry out recurrence to each closed contour using Ramer algorithms Point, until the ultimate range of obtained whole line segments to each self-corresponding closed contour is less than threshold value Dmax；Again to straightway and/ Or arc is fitted with merging, the intersection point of adjacent segments is calculated according to the linear equation and/or elliptic arc equation fitted Coordinate, the intersection point have corresponded to the target two-dimensional silhouette in step 2；

Step 4, profile is specified search for, caused all closed contour confirmations are identical with To Template profile in step 3 for retrieval The profile undetermined of line segment composition, treat the wide progress posture computing of fixed wheel and complete to confirm, in the picture the successful mesh of marker recognition Mark, export target location and projection matrix.

2. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 2, the ambiguity profile is in the sequentially description of the profile, identical be present and sequentially describes；Establish target two Dimension profile geometric primitive topological relation specific method be

Primitive information is parsed from the ENTITIES sections for the file for including image outline information, includes the start-stop of Nl bar straight lines Coordinate, the center of circle of Na bar circular arcs, radius, start-stop coordinate, centre coordinate, the radius of Nc circle, preserve all straight lines, circular arc and circle Information；

Closed contour is searched, since the starting point of any straight line or circular arc, lookup overlaps next with the straight line terminal The starting point of connecting line segment, and this process is repeated, until returning to the starting point most started, that is, find a closed contour；

The line segment included in the closed contour found is rejected from former data, the step of repeating to search closed contour, directly Finished to all line segments are searched, count N number of closed contour, plus the circular Nc in objective contour being originally self-closing profile, Amounting in target two dimensional model has N+Nc closed contour, counts each closed contour cathetus number Nl_i, Circular Arc Na_i, circle Number Nc_i, the numbering of wherein footmark i expression closed contours；

Sequentially relation, description rule are from any line for the connection of the straight line and camber line of non-circular closed contour in description closed contour Section starts, and returns to starting point along the fixed all line segments for traveling through profile clockwise or counterclockwise, is then designated as 0 if straight line, arc Line is designated as 1；For each non-circular closed contour, sequentially relation shares (NL for it_i+NA_i) kind, i.e., using each line segment as Point, obtain a kind of connection sequentially, complete the foundation of topological relation.

3. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 2, its feature exist In：In the step 2, in addition to it is the checking step that ambiguity profile is also non-ambiguity profile to closed contour；Ambiguity profile is In the sequentially description of the profile, identical be present and sequentially describe.

4. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 3, the method for excluding non-closed profile is：All marginal points are traveled through, if some marginal point is thirdly neighborhood In, the quantity of other marginal points is only 1, then the marginal point belongs to patent chalaza, after deleting the marginal point, is reexamined and patent The adjacent marginal point of chalaza, thirdly in neighborhood whether also only have other marginal points, one can be deleted by that analogy not The profile of closure；Said process is repeated, after having handled all marginal points, remaining profile is then all closed contour.

5. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 3, the method to closed contour single-pass, more logical marginal point inspections is the five neighborhood outermosts one at a certain edge In circle, the quantity of non-adjacent marginal point is checked, being checked whether for the non-adjacent marginal point in everywhere can be along thirdly neighborhood reaches Center edge point, if can if show at this centered on marginal point a passage, if can not if at this it is unrelated with center edge point, Directly exclude, be otherwise single channel edge for multi-channel edge point if the number of channels of the center edge point is more than 2 finally Point.

6. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 3, intersecting outlines separation method is single-pass, the more logical marginal point to checking, since any marginal point, If the marginal point is single-pass marginal point, the next marginal point of an optional search channels, if more logical marginal points, then edge respectively Each next marginal point of search channels；If having returned to the marginal point searched in search procedure, the route is deleted, directly Starting point is returned to search, more all paths for successfully returning to starting point, takes most short one to be used as minimal closure profile, in repetition Process is stated, completes the separation to all intersecting outlines.

7. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 3, straightway and/or arc are fitted and to merge specific method as follows：

Oval segmental arc fitting, checks the adjacent both sides for the polygon for approaching closed contour, uses the side where ellipse fitting both sides Edge point, ellipse fitting error is calculated, if error is less than the approximate error of both sides and correspondence profile, two are replaced using elliptic arc Straight line, said process is repeated along adjacent two edges, until can not be preferably fitted using elliptic arc；

Oval segmental arc merges, if carrying out continuous multi-stage occur in recursion to each closed contour using Ramer algorithms The profile being fitted using elliptic arc, the center spacing of two adjacent ellipses arcs is calculated, being represented if threshold value Dcmax is less than can To merge, after merging into an elliptic arc, repeat and carry out same operation with adjacent next elliptic arc, until adjacent ellipse Circular arc is disposed；

Fitting a straight line carries out that in recursion elliptic arc can not be used with merging, using Ramer algorithms to each closed contour The profile being fitted then directly is fitted using straight line, and carries out judgement merging to adjacent straightway；

The intersecting point coordinate of adjacent segments is sought, is AX according to linear equation Y=kX+b and elliptic arc equation²+BXY+CY²+DX+EY+1 =0 solves the intersecting point coordinate (X, Y) of adjacent segments, and the intersection point corresponds to the node of To Template；K, b are what the Fitting Calculation went out in formula Coefficient, only straight line is characterized, and (A, B, C, D, E) is the parameter that ellipse fitting calculates, each group of parameter characterization is uniquely ellipse Circle.

8. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 4, the method for specifying search for profile is, if multiple closed contours in objective contour be present, preferred non-ambiguity Profile is as template contours, most from line segment number using the most profile of line segment number as template contours if being all ambiguity profile Profile along angle point, all calculate similarities, select most like one group.

9. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In：In the step 4, the specific method for confirming the profile undetermined of line segment composition identical with To Template profile is to retrieve in step Caused all closed contours in 3, reject the composition line segment number profile different from template contours line segment；In remaining profile, appoint Meaning selects a node in the profile, the connection of the profile is arranged in the direction of the clock sequentially, if this sequentially can be in template Found in (NLi+NAi) of profile in connection sequentially, then the search profile is profile undetermined, is otherwise directly excluded.

10. a kind of identification and space-location method based on target two-dimensional silhouette model as claimed in claim 1, its feature exist In in the step 4, the method that profile undetermined carries out posture computing and completes to confirm is following steps：

According to profile undetermined and the sequentially relation of template contours, the corresponding relation of physics seat and pixel coordinate between node is determined, Template contours provide the physical coordinates of node, and the node of profile undetermined is pixel coordinate, initially sets up Attitude Calculation equation

Optional three not conllinear nodes in template contours, by its physical coordinates [X_w Y_w 0]_i ^TAnd detected in profile undetermined The pixel coordinate [u0 v0] gone out_i ^TBring above-mentioned Attitude Calculation equation into, obtain that projective transformation matrix is calculated

Projective transformation matrix is brought into above-mentioned Attitude Calculation equation, by all node physical coordinates [X of template contours_w Y_w 0]_i ^TPoint Do not bring above-mentioned Attitude Calculation equation into, calculate the projected pixel coordinate [u v] of each node_i ^T；

Projection similarity F is calculated, according to the pixel of the corresponding node detected in node pixel coordinate (u v) i and profile undetermined Coordinate (u0v0) i calculates similarity F, and calculation formula is as follows

Wherein n represents nodal point number summation in template contours；

For non-ambiguity profile, if similarity exceedes given threshold, then it is assumed that the profile undetermined is correct target, and is returned to Projective transformation squareBattle array, the matrix then describe spatial translation position of the target with respect to camera and its around The anglec of rotation of each axle；

For ambiguity profile, if the similarity under a variety of node corresponding relations exceedes given threshold, multiple postures will be obtained Matrix, then the profile undetermined is set to polymorphic profile, if target only has this ambiguity profile, directly returned satisfactory Multiple attitude matrixs；

If the common attitude matrix of existence anduniquess between multiple polymorphic profiles, the attitude matrix is the correct posture of ambiguity profile； If still there is multiple identical attitude matrixs, the distribution that illustrates multiple ambiguity profiles of target be it is full symmetric, then directly Return to multiple attitude matrixs, and the successful target of marker recognition in the picture.