CN104252701B - Correct the method and system of disparity map - Google Patents
Correct the method and system of disparity map Download PDFInfo
- Publication number
- CN104252701B CN104252701B CN201310269280.1A CN201310269280A CN104252701B CN 104252701 B CN104252701 B CN 104252701B CN 201310269280 A CN201310269280 A CN 201310269280A CN 104252701 B CN104252701 B CN 104252701B
- Authority
- CN
- China
- Prior art keywords
- edge
- disparity map
- matrix
- edges
- degree
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Abstract
Open bearing calibration and system there is provided a kind of disparity map.This method includes：Detecting step, detects one or more first edges of the first disparity map and one or more second edges of reference picture；And aligning step, based on the matching of the one or more first edges and one or more second edges detected, first disparity map is corrected with the second disparity map after being corrected.
Description
Technical field
Present application relates generally to correct the method and system of disparity map, and more particularly, to based on anaglyph and ginseng
The distribution at the edge in image is examined to correct the method and system of disparity map.
Background technology
Stereovision technique is widely used to three-dimensional movie, virtual reality, automated vehicle control, Study of Intelligent Robot Control
Deng numerous areas.
With the rise in the fields such as three-dimensional reconstruction, virtual reality, precision of the people to the disparity map in stereoscopic vision
With the requirement more and more higher of density, expect to obtain the reliable depth information of each pixel in image.Stereo matching
Purpose is exactly that the pixel pair of matching is found in two width imaging planes of identical three-dimensional scenic, calculates each pixel point coordinates
Corresponding parallax value d, and then the three-dimensional spatial information of scene can be described exactly.But existing technology is obtained in the prior art
To disparity map it is not accurate enough, it is necessary to be improved, improve parallax precision.
The patent application No.US8,300,085B2 as disclosed in YANG QIONG et al. 30 days October in 2012 is proposed
One occlusion issue in Stereo matching is handled based on image segmentation algorithm.The left side that this method is shot to binocular camera first
Figure and right figure carry out image segmentation, and the result split based on image finds the border between two segmentation blocks, so as to find blocked area
Domain, the parallax value of image pixel is obtained followed by an energy equation optimization.However, this method needs to do multiple image point
Cut, and split the border found using image and be likely to be mistake, may not necessarily block caused.In addition, this
The method of kind individually considers border, ignores the relation between border, therefore can not obtain gratifying effect under many circumstances
Really.
Ramya Narasimha、Elise Arnaud、Florence Forbes、Radu Horaud:Cooperative
disparity and object boundary estimation.ICIP2008:1784-1787 bibliography proposes a kind of
The algorithm that comprehensive disparity computation and object boundary estimation is integrated, this method is using unified Markov framework two tasks
Unite, propose a joint ensemble, parallax is estimated using markov random file.But this method is for figure
The estimation of each pixel parallax of picture, one side amount of calculation is larger, on the other hand, will certainly band from single pixel pointwise estimation
Carry out noise, can not also obtain gratifying disparity map.
The content of the invention
In view of above mentioned problem, present disclose provides a kind of bearing calibration of disparity map and system.
According to an aspect of this disclosure there is provided a kind of method for correcting disparity map, including：Detecting step, detection first
One or more first edges of disparity map and one or more second edges of reference picture；And aligning step, based on institute
The matching of the one or more first edges and one or more second edges of detection, corrects first disparity map to obtain school
The second disparity map after just.
According to another aspect of the present disclosure there is provided a kind of system for correcting disparity map, including detection means, inspection is configured as
Survey one or more first edges of the first disparity map and one or more second edges of reference picture；And means for correcting,
It is configured as the matching based on the one or more first edges and one or more second edges detected, correction described first
Disparity map is with the second disparity map after being corrected.
Brief description of the drawings
Exemplified with embodiments of the invention in the figure of the reference of accompanying drawing, in accompanying drawing, to the description of figure middle identical in the whole text
Numeral refers to identical key element.
Fig. 1 shows that the application scenarios during method using correction disparity map according to an embodiment of the invention show
Example.
Fig. 2 shows the example flow diagram of the method for correction disparity map according to an embodiment of the invention.
Fig. 3 shows the effect example during method of the correction disparity map shown in application drawing 1.
Fig. 4 shows the detailed example flow of the method for correction disparity map according to another embodiment of the invention
Figure.
Fig. 5 (a) and (b) are shown in the method for correction disparity map according to an embodiment of the invention according to initial parallax figure
Set up the schematic process of preliminary matches degree matrix.
Fig. 6 (a) and (b) are shown in the method for correction disparity map according to an embodiment of the invention according to initial parallax figure
Set up the schematic process of similarity matrix.
Fig. 7 shows the block diagram of the system of correction disparity map according to another embodiment of the present invention.
Embodiment
It is described in detail with reference to specific embodiment and accompanying drawing.
Fig. 1 shows that the application scenarios during method using correction disparity map according to an embodiment of the invention show
Example.
In Fig. 1, device 100 is binocular stereo camera, and image 102 (1) and 102 (2) are that binocular camera is defeated respectively
The two matching images gone out（Also commonly referred to as left/right image, is gray-scale map in this example embodiment）；Image 103 is to be based on matching figure
Picture（Left/right image）An obtained disparity map；Then, the disparity map 103 and two matching one of images（Such as left figure
As 102 (1)）It is input into the system 101 for correcting disparity map（Or perform software, hardware, firmware, the chip of method etc.
Deng）, this can be realized by chip 104, then, the disparity map 105 after being corrected.
Fig. 2 shows the example flow diagram of the method 200 of correction disparity map according to an embodiment of the invention.This method
200 can be the method 101 run in the scene of Fig. 1 examples by chip 104.
The method 200 of the correction disparity map includes：Detecting step S201, one or more the first of the first disparity map of detection
One or more second edges at edge and reference picture；And aligning step S202, based on one or more detected
The matching at one edge and one or more second edges, corrects first disparity map with the second disparity map after being corrected.
In this embodiment, the example of first disparity map can be an initial parallax figure.I.e..The correction of disparity map can
To be carried out based on an initial parallax figure.Certainly, the effect of the initial parallax figure equally can also influence changing for disparity map
Kind precision.The initial disparity map can show that such as Block- matching, dynamic are advised using the method for various traditional calculating parallax values
The method of drawing, figure cut method, half global registration method（Semi-Global Matching, SGM）Etc..Certainly, should the invention is not restricted to this
First disparity map can also be the disparity map after initial parallax figure or correction（Further to be corrected）Or other are any
Disparity map.The disclosure is not intended to limit to first disparity map.
The reference picture can be the left image or right image or other kinds of reference that binocular camera is shot
Image, if its can for the captured scene of description edge feature.
In this method 200, detect the first disparity map one or more first edges and one or many of reference picture
The mode of individual second edge can also use the mode of traditional rim detection, for example, the biography based on search and based on zero crossing
System edge detection method.The details for the rim detection that will not be described here.
The proposition of correction disparity map in the present embodiment can be based on two hypothesis：1）The hop region one of initial parallax
As be real parallax discontinuity zone；2）Real parallax discontinuity zone is typically in border or the border land of target object
Side.Based on above-mentioned two it is assumed that can by cause the border or edge of the detection in initial parallax figure with（Generally can be from bat
Obtained in the gray-scale map taken the photograph）Target object border or edge are consistent, to correct initial parallax value.For this purpose, originally
Embodiment can go to correct initial parallax value using the edge of reference picture, briefly, can be with for every in initial parallax figure
A line edge, selects the edge of a matching consistent to be modified to, to correct the initial parallax figure in a reference image.
In one embodiment, at least one in one or more of first edges and one or more second edges
Length can be more than predetermined length threshold value.That is, can be regarded for the first of reference picture and such as initial parallax figure
Difference figure extracts each edge respectively, then filters short edge, only retains long edge（For example, length is more than predetermined length threshold value
Edge）.The selection of predetermined length threshold value is general related with actual demand, if for example, the scene bag of to be improvedd disparity map
Relatively large artificial structure, such as high building, electric pole are included, threshold value typically chooses larger, typically more than such as 15 pixels；
If the scene of to be improvedd disparity map includes small-sized object, such as automobile, threshold value is typically chosen more than 8-9 pixel.Certainly,
The selection of these threshold values is only example and unrestricted.Retaining the benefit at long edge can include：Reduce the calculating of matching short edge
Amount, meanwhile, matched using long edge, because the matching at long edge less easily malfunctions, therefore matching can make it that more
Plus it is accurate.
Sometimes for actual demand, it is necessary to handle long but also straight edge not only emphatically（Because being easier using long straight edge
More accurately matched）, so in practice, may be except needing the length at calculating edge, in addition it is also necessary to consider the curved of edge
Qu Chengdu, here the degree of crook at edge weighed with bent curvature of a curve.On the meaning of curvature, reference can be made to webpagehttp:// baike.baidu.com/view/562504.htm.I.e. the length at edge is more than predetermined length threshold value and the curvature at edge is less than
The edge of predetermined curvature threshold value may be considered long straight edge.And then, more accurately corrected by matching these long straight edges
Disparity map.
In the disclosure, only it is described by taking long edge as an example, but this is not limitation.
In this method 200, the aligning step S202 can include:Step S2021 (not shown), based on what is detected
The size and location of one or more first edges and one or more second edges, determine each first edge with each second
The matching degree at edge；In step S2022 (not shown), the matching degree based on specific first edge and all second edges
Maximum one, it is determined that the corresponding second edge matched with each first edge；Step S2023 (not shown), will be described each
It is consistent that individual first edge is modified to its corresponding second edge；And step S2024 (not shown), based on it is revised each
First edge, to correct first disparity map.
No matter in addition, the matching of long edge or short edge can use various known methods, for example, traditional image
By the coefficient correlation of the window of formed objects in comparison object area and the field of search in the matching of method of completing the square, such as two dimensional image, take
Window center point in the field of search corresponding to coefficient correlation maximum is used as match point etc..The present inventor's design explained below
A kind of matching way at novel edge.
It is described based on the one or more first edges detected and one or more second edges in method 200
Size and location, determining the step S2021 of each first edge and the matching degree of each second edge can include：
For each specific first edge vi：Select and the specific first edge v in a reference imageiFirst
There are one or more specific second edges of coincidence on direction；Calculate the specific first edge viWith the one of the selection
The distance in the second direction different from first direction of individual or multiple specific second edges, is used as described specific first
Each preliminary matches degree of edge respectively with one or more of specific second edges；And
Set up preliminary matches degree matrix Y(m×n){yij, m and n are the number of the first edge and second edge respectively here
Mesh, wherein the preliminary matches degree matrix Y(m×n){yijIn element yijIt is i-th of first edge and reference in the first disparity map
The preliminary matches degree of j-th of second edge in image, wherein, i is positive integer, and i<=m, and j is positive integer, and j<=n.
The first direction can be short transverse, and the second direction can be width.But, the invention is not restricted to
This, first direction and second direction can also take other directions.
In fact, passing through preliminary matches degree matrix Y(m×n)Each first side in the first disparity map can have been obtained
Edge and the preliminary matches degree of each second edge in reference picture, can be as the matching degree in step S2021
.In step S2022, using these preliminary matches degree, it can learn which first edge is most matched with which second edge
（That is, which is maximum in all preliminary matches degree by comparing a specific first edge and all second edges, or compares
It is more than predetermined matching degree threshold value in all preliminary matches degree）, then treat them as matching each other.So as in step
In S2023, each described first edge is modified to its corresponding second edge consistent；And in step S2024, based on repairing
Each first edge after just, to correct first disparity map.
Therefore, above-mentioned preliminary matches degree matrix is utilized, it is already possible to correct initial parallax figure accurately to be regarded
Difference figure, so as to be laid the foundation based on the Stereo matching of disparity map or other stereoscopic vision applications in the future.
But, due to the preliminary matches degree be only each first edge for being related in the first disparity map with reference picture
Distance between each second edge in the direction of the width, therefore, if each first edge for being obtained by rim detection or
There is the edge of error detection in each second edge, for example, being originally a line edge of object, by wrong after rim detection
Ground is detected as the close edge in two positions, or is two edges of two objects originally, is examined by ground wrong after rim detection
Survey as a line edge, in this case, it is possible to the edge of jobbie should be matched with a certain second edge originally, but by
In being mistakenly detected as two edges, then it is one and a certain second edge in this two edges that may cause matching result
Matching, and another matches with another second edge.Accordingly, it may be desirable to further change the preliminary matches degree matrix.
It is described based on the one or more first edges detected and one or more for example, in one embodiment
The size and location at two edges, determines that the step S2021 of the matching degree of each first edge and each second edge can be with
Including：The distribution of second edge in distribution character and the reference picture based on the first edge in first disparity map
At least one in characteristic, changes the preliminary matches degree matrix, to obtain amended matching degree.
If in the distribution character and the reference picture by analyzing the first edge in first disparity map
At least one in the distribution character at two edges, it may should be one be mutually associated that can learn which the first/bis- edge
Edge, or it be two or more edges which the first/bis- edge, which is probably, so as to further improve each first edge
With the matching relationship of each second edge.
In one embodiment, the distribution character of the first edge in first disparity map can be obtained by following steps
Arrive：For each two first edge Vi′And Vj′Calculate their central point cpi′And cpj′Minimum range in a first direction；Such as
Really the minimum range on the first direction is less than a distance threshold, then calculates central point cpi′And cpj′In the distance of second direction,
It is used as described two first edge vi′And vj′Similarity；Set up similarity matrix W(m×m){wi′i′Be used to change described preliminary
Matching degree matrix, m is the number at the edge in the first disparity map, element w therein herei′j′It is the i-th ' individual in the first disparity map
The similarity degree at edge and jth ' individual edge, wherein i ', j ' are positive integers, and i '<=m and j '<=m.Here, due to similarity moment
Battle array is set up according to the distance between two edges in the first disparity map, according to from must be nearer more have can at two edges
Can be same edge, therefore point of first edge that the similarity matrix can reflect in the first disparity map to a certain extent
Cloth characteristic, for example, which first edge may should be an edge being mutually associated, or which first edge is probably to answer
This is two or more edges.
In another embodiment, the distribution character of the second edge in the reference picture can be obtained by following steps
Arrive：For each two second edge ui′′And uj′′, calculate their central point cpi′′And cpj′′Most narrow spacing in a first direction
From；If the minimum range on the first direction is less than a distance threshold, central point cp is calculatedi′′And cpj′′In second direction
Distance, be used as described two second edge ui′′And uj′′Consistent degree；Set up consistent degree matrix X(n×n){xi′′j′′Be used to repair
Change the preliminary matches degree matrix, n is the number at the edge in reference picture, element x therein herei‘’j‘’It is reference picture
In the i-th ' ' individual edge and jth ' ' individual edge the degree of consistency, wherein i ' ', j ' ' be positive integer, and i ' '<=n and j ' '<=
n.Here, because consistent degree matrix is set up according to the distance between two edges in reference picture, according to from must get over
Two near edges are more likely to be same edge, therefore the consistent degree matrix can reflect reference picture to a certain extent
In second edge distribution character, for example, which second edge may should be an edge being mutually associated, or which
It is two or more edges that second edge, which is probably,.
It is above-mentioned, the distribution character of the distribution character of the first edge of the first disparity map and the second edge of reference picture
Only it is example to method, it is contemplated that other modes obtain these distribution characters, as long as each edge can be embodied in space
On position and size between relation.
The above-mentioned modification to preliminary matches degree can be based on the first edge of above-mentioned first disparity map distribution character and ginseng
One in the distribution character for the second edge for examining image, or based on both.Preliminary matches degree is being changed based on both
In the case of, more accurately amended matching degree can be obtained.
In one embodiment, the distribution character of the first edge based in first disparity map and the reference
At least one in the distribution character of second edge in image, changes the preliminary matches degree matrix, amended to obtain
The step of matching degree, can also include：Distribution based on the edge in preliminary matches degree matrix and the initial parallax figure
At least one in the distribution character of characteristic and the second edge in the reference picture, builds energy equation；Minimize described
Energy equation, draws the amended matching degree at each edge in initial parallax figure；Build revised matching degree matrix
F(m×n), every a line of the matrix is the amended of each edge of an edge in initial parallax figure and reference picture
Matching degree.
In one embodiment, the distribution at the edge in based on preliminary matches degree matrix and the initial parallax figure
Both distribution characters of second edge in characteristic and the reference picture are built during the energy equation, can by as follows
Formula is built：
E(F)=El(F)+αEs(F)+βEc(F),
Wherein, El(F)=tr{(F-Y)T(F-Y) }, Es(F)=tr{FTLF }, Ec(F)=tr{FCFT, α and β are constant, its
In, L=D-1/2ΔD-1/2, wherein Δ=D-W, D is a diagonal matrix, each diagonal element be equal to each row element of matrix W it
With；Wherein, C=D '-1/2Δ′D′-1/2, wherein Δ '=D '-X, D ' is a diagonal matrix, and each diagonal element is equal to matrix X
Each row element sum.Here, tr { } is the mark of matrix, if F is similar with Y, El(F) value is just smaller.The energy side
Journey is actually the similarity matrix W and reference picture that consider the first disparity map consistent degree matrix X to preliminary matches degree matrix Y
Influence.
Then, the energy equation E (F) is minimized so that obtain matching degree matrix F when taking what value, E (F) minimums, from
And amended matching degree matrix F is drawn, so as to draw the amended matching degree at each edge in initial parallax figure；Structure
Build revised matching degree matrix F(m×n), every a line of the matrix is an edge and reference picture in initial parallax figure
The amended matching degree at each edge.
In this way, can be in the hope of amended matching degree matrix F by above-mentioned formula(m×n), every a line of the matrix is initial
The amended matching degree at each edge at an edge and reference picture in disparity map.So as to continue step S2022
Maximum one in (not shown), the matching degree based on specific first edge and all second edges, it is determined that with each the
The corresponding second edge of one edge matching；Step S2023 (not shown), each described first edge is modified to corresponding to its
Second edge it is consistent；And step S2024 (not shown), based on each revised first edge, to correct described first
Disparity map.
Therefore, using above-mentioned amended matching degree matrix, more accurate disparity map can be obtained, so that in the future
More preferable basis is established based on the Stereo matching of disparity map or other stereoscopic vision applications.
Fig. 3 shows the effect example during method of the correction disparity map shown in application drawing 1.
It can be seen from figure 3 that first, being extracted respectively to reference picture and initial parallax figure（Or detection）Edge, and preferably, mistake
Short edge is filtered, only retains long edge.So as to using the edge with the edge matching in initial parallax figure of reference picture, to instruct
And each edge of initial parallax figure is corrected, improve initial parallax figure with this.Specific reference picture as can see from Figure 3
With the example of initial parallax figure, the example at the edge of specific reference picture and the edge of initial parallax figure, and how to filter
Short edge, retains the example at long edge, and how to be corrected using the edge of the matching in reference picture in initial parallax figure
Edge, come the example of the disparity map after being improved.
From Fig. 3 it can also be seen that according to the disparity map that can be improved of each embodiment of the present invention, so that in the future
Laid the foundation based on the Stereo matching of disparity map or other stereoscopic vision applications.
Fig. 4 shows the detailed exemplary flow of the method 400 of correction disparity map according to another embodiment of the invention
Cheng Tu.
In Fig. 4, first, input be reference picture 401 and initial parallax Figure 40 2 and their rim detections result
403.Then, alternatively, by calculating the length at edge（For example, the number of the pixel at edge）With the comparison of length threshold,
Short edge filtering 404, so as to only remain long edge.Then, preliminary of edge in initial parallax figure and reference picture is calculated
With degree matrix 4011.Alternatively, the similarity matrix 4012 at initial parallax figure edge is calculated.Alternatively, calculate in reference picture
The consistent degree matrix 4013 at edge.Above-mentioned preliminary matches degree, similarity and consistent degree are preferably based on, energy equation is minimized,
And calculate the matching degree at amended each edge（Fraction）405, based on above-mentioned matching degree, draw initial parallax figure
Each of the edges, corresponding to the edges matched of reference picture, is then based on each edge of matching, draws the disparity map after correction
406。
Specifically, the calculating of preliminary matching degree and initial matching degree matrix Y structure 4011 are referred to shown in Fig. 5, Fig. 5
Show to set up preliminary matches degree matrix according to initial parallax figure in the method for correction disparity map according to an embodiment of the invention
Schematic process.
Specifically, in Figure 5, with vi(i=0,1 ..., m) and uj(j=0,1 ..., n) come represent respectively initial parallax figure and
The first edge and second edge of the length as example in reference picture, wherein i and j are initial parallax figure and with reference to figure respectively
The call number at the long edge as in, m and n are initial parallax figure and the number with reference to long edge in figure.For initial parallax figure
Edge v each longi, main step is for example：
1）Calculate long edge viCentral point cpi, any center point calculating method can be selected here, for example, calculate line segment
Center of gravity, center etc..
In following step, the direction at edge can also be calculated.Preferably, when two edges to be calculated and height side
During to angle in less than 45 degree（I.e., it is possible to just for edge approximately on short transverse Y, and ignore on width X
Edge）, calculated by following step；Otherwise（That is, most of is edge on width）, Y-coordinate（The seat of short transverse
Mark）Can and X-coordinate（Width）Exchange.
Herein, it is preferable that be only the reason for the matching at the consideration approximate edge on short transverse Y, due to binocular
Video camera is usually to be set parallel to ground, i.e., set parallel to width, therefore the disparity map and reference picture of its generation
The edge of gray-scale map in the direction of the width will not generally produce skew, therefore, it can the edge of emphasis amendment in the height direction
Skew.Certainly, the invention is not restricted to this, if in case of need, binocular camera set perpendicular to ground, i.e.,
When being set parallel to short transverse, then can then may be used by the way that the short transverse Y and width X in the disclosure are interchangeable
To adapt to such case.Here, not repeating.
2）In a reference image, selection and the first long edge viIn the Y direction（The example of first direction）There is the long side of coincidence
Edge uj；
3）Calculate the specific first edge viWith one or more specific second edge u of the selectionjWith
Distance in the different second direction of first direction.For example, the first long edge v can be calculatediCentral point cpiWith select
Long edge u in the reference picture comejBetween in X-direction（The example of second direction）Distance（This can be by asking first
Long edge viCentral point cpiWith the long edge u in the reference picture chosenjCentral point cpjBetween in the X direction
Distance）, such as Fig. 5（a）It is shown.Certainly this asks the mode of distance to be only example using central point, it is also contemplated that other modes are come
Seek distance.
4）Build preliminary matches degree matrix Y(m×n){yij, during m and n are respectively initial parallax figure and reference picture here
The number at edge, element y thereinijIt is the preliminary of j-th strip edge in i-th edge and reference picture in initial parallax figure
Matching degree, its definition can be equation below：
Here dx(vi,uj) represent edge viAnd ujThe distance in X-direction, σyIt is the parameter in Gauss model, in order to control
D processedx(vi,uj) value degree of scatter.Certainly, formula yijLimitation is only exemplary rather than, it mainly represents edge viAnd uj's
The influence brought in the distance of X-direction.
In this way, preliminary matches degree matrix can be obtained.
In fact, passing through preliminary matches degree matrix Y(m×n)Each first side in the first disparity map can have been obtained
Edge and the preliminary matches degree of each second edge in reference picture, can be as the matching degree in step S2021
.In step S2022, using these preliminary matches degree, it can learn which first edge is most matched with which second edge
（That is, which is maximum in all preliminary matches degree by comparing a specific first edge and all second edges, or compares
It is more than predetermined matching degree threshold value in all preliminary matches degree）, then treat them as matching each other.So as in step
In S2023, each described first edge is modified to its corresponding second edge consistent；And in step S2024, based on repairing
Each first edge after just, to correct first disparity map.
But, due to the preliminary matches degree be only each first edge for being related in the first disparity map with reference picture
Distance between each second edge in the direction of the width, therefore, if each first edge for being obtained by rim detection or
There is the edge of error detection in each second edge, in this case, it is possible to cause the matching result of mistake.Therefore, may be used
It can need further to change the preliminary matches degree matrix.
Citing below but without limitation, the distribution character of first edge that how to obtain in the first disparity map and described
The specific example of the distribution character of second edge in reference picture.
Fig. 6 shows to set up similar according to initial parallax figure in the method for correction disparity map according to an embodiment of the invention
Matrix is spent as the schematic process of an example of the distribution character of the first edge in the first disparity map.
Similarity in initial parallax figure between each edge can be referred to simply between each edge in width side
Upward distance is weighed.This is to consider, two in the direction of the width close edge be likely to actually should be a line
Edge.
Based on this, the calculating of the similarity in initial parallax figure between each edge can be as shown in Figure 6, it is assumed that vi′With
vj′For two long edges in initial parallax figure, the key step of example is as follows：
1）Calculate the central point cp at two long edgesi′And cpj′；
In following step, the direction at edge can be calculated.As it was previously stated, only when two edges to be calculated and height
Spend direction（The example of first direction）In the angle below 45 degree（That is, illustrate that these edges are almost vertical）When, just press
The step of face, calculates；Otherwise, Y-coordinate can be replaced with X-coordinate.
Herein, it is preferable that be only the reason for the matching at the consideration approximate edge on short transverse Y, due to binocular
Video camera is usually to be set parallel to ground, i.e., set parallel to width, therefore its disparity map produced is in width
On edge will not generally produce skew, therefore, it can the skew at the edge of emphasis amendment in the height direction.Certainly, this hair
Bright not limited to this, if in case of need, binocular camera set perpendicular to ground, i.e., sets parallel to short transverse
When putting, then such case can be then adapted to by the way that the short transverse Y and width X in the disclosure are interchangeable.
This is not repeated.
2）Calculate vi′And vj′In the Y direction（Short transverse）Minimum distance, if the distance is less than a threshold value, is counted
Calculate the X-direction of two central points（Width, is used as the example of second direction）Shown in distance, such as Fig. 6 (a).In this way, by first
The distance of the central point at two edges of beginning disparity map in the direction of the width simply represents the similarity at the two edges.Phase
It is higher like spending, illustrate that the two edges are more possible to actually same edge.
3）Build similarity matrix W(m×m){wi′j′, m is the number at edge in initial parallax figure, element therein here
wi′j′It is the similarity at the i-th ' bar edge and jth ' bar edge in initial parallax figure, it is defined as follows：
Here dx(vi′, vj′) represent edge vi′And vj′X-direction distance, σwIt is the parameter in Gauss model.The formula
wi′j′Limitation is only exemplary rather than, it mainly represents edge vi′And vj′The influence brought in the distance of X-direction.
Simply initial parallax figure is weighed shown in certain Fig. 6 with reference to the distance between each edge in the direction of the width
In the mode of similarity between each edge be only example, other modes can be conceived to embody each edge of initial parallax figure
Between similarity, to avoid a line edge in practice being mistakenly detected as two edges.
Similarly, it is also possible to exist in a reference image that a line edge in practice is mistakenly detected as into two edges,
Or by this kind of situation for being detected as a line edge of two marginal errors in practice, therefore can by with above-mentioned initial
The Similarity Measure carried out in disparity map similarly, calculates the consistent degree between the edge in reference picture.Similarly, the consistent degree
It is higher, illustrate that the two edges are more possible to actually same edge.
The calculating of the consistent degree at the edge in reference picture（Not shown in figure）With the similarity at edge in initial parallax figure
Calculating it is closely similar.Set ui′′And uj′′For with reference to two long edges in figure,
1）Calculate the central point cp at two long edgesi′′And cpj′′；
In following step, the direction at edge can be calculated.As it was previously stated, only when two edges to be calculated and height
Spend direction（The example of first direction）In the angle below 45 degree（That is, illustrate that these edges are almost vertical）When, just press
The step of face, calculates；Otherwise, Y-coordinate can be replaced with X-coordinate.
2）Calculate ui′′And uj′′In the Y direction（Short transverse）Minimum distance, if the distance is less than a threshold value, is counted
Calculate the X-direction of two central points（Width, is used as the example of second direction）Shown in distance, such as Fig. 6 (a).In this way, by first
The distance of the central point at two edges of beginning disparity map in the direction of the width simply represents the similarity at the two edges.Phase
It is higher like spending, illustrate that the two edges are more possible to actually same edge.
3) consistent degree matrix X is built(n×n){xi′′j′′, element x thereini‘’j‘’Be with reference to the i-th ' in figure ' article edge and
The consistent degree at j ' ' bars edge, it is defined as follows：
Here dx(ui′′,uj′′) represent edge ui′′And uj′′X-direction distance, σcIt is the parameter in Gauss model.The dx
(ui′′,uj′′) limitation is only exemplary rather than, it mainly represents edge ui′′And uj′′The influence brought in the distance of X-direction.
In this way, can further embody the edge of the detection in initial parallax figure by the similarity matrix of initial parallax figure
Between similarity, i.e. whether they are probably actual same edge.And can by reference to the consistent degree matrix of image
The similarity between edge further to embody the detection in reference picture, i.e. whether they are probably actual same
Bar edge.It therefore, it can consider at least one in the similarity matrix of initial parallax figure and the consistent degree matrix of reference picture
（I.e., single similarity matrix or individually consistent degree matrix, or both）Come calculate before further changing obtain just
Matching degree matrix between the edge of beginning disparity map and the edge of reference picture.
It is given below and is based only upon the similarity matrix of initial parallax figure to change the example of preliminary matches degree matrix F.
Energy equation（Or loss equation）It may be defined as：E(F)=El(F)+αEs(F).The definition of the energy equation is also
It is known in the art that those skilled in the art can construct other different from the formula disclosed in the disclosure according to its principle
Formula.
In the disclosure,
El(F)=tr{(F-Y)T(F-Y)}
Es(F)=tr{FTLF}
Δ=D-W,L=D-1/2ΔD-1/2
E(F)=tr{(E-Y)T(F-Y)}+αtr{FTLF}
Target equation is：F=argmin{E(F)}.Here, the argmin { E (F) } seeks the meaning of E (F) minimum value.α is
Constant parameter.
Partial differential is asked with F to above formula E (F), and causes the result after partial differential to be equal to 0（That is, E when what value F take is calculated（F）
Take minimum value）, can obtain
(αL+I)F=Y
It is 1 on diagonal that I, which is, and other are 0 unit matrix.
When α L+I can the inverse time, have unique solution, i.e. F=(α L+I)-1Y。
Revised matching degree matrix F is obtained in this way, can calculate.
Below, both consistent degree matrixes of the similarity matrix for being based only upon initial parallax figure and reference picture are provided to change
The example of preliminary matches degree matrix F.
Namely based on above-mentioned initial matching degree, similarity and consistent degree three, following energy equation is minimized, it is every to solve
The matching degree at one long edge（Fraction）, wherein energy equation in this case can be defined as follows：
E(F)=El(F)+αEs(F)+βEc(F)
Here loss equation El(F) it is, that in order to consistent with " matching assume ", that is, the matching degree fraction changed must be with
Initial matching degree（Fraction）There are certain association, initial matching degree（Fraction）Even if it is not accurate enough, but also have certain finger
Lead effect.
Represented with matrix form, El(F) it can be defined as follows：
El(F)=tr{(F-Y)T(F-Y)}
Here tr { } is the mark of matrix, if F is similar with Y, El(F) value is just smaller.
Loss equation Es(F) it is a regularization term, in order to, i.e., in initial parallax with " similarity hypothesis " unanimously
Matching degree of the figure at a distance of closer edge should be more similar.We are come using the similarity matrix W of above-mentioned initial parallax figure
The similarity at edge in initial parallax figure is evaluated, W is normalized, L=D is obtained-1/2ΔD-1/2, wherein Δ=D-W, D is one diagonal
Matrix, each diagonal element is equal to each row element sum of matrix W.
So as to loss equation Es(F) it is defined as：
Es(F)=tr{FTLF}
With Es(F) it is similar, loss equation Ec(F) it is also a regularization term, it is therefore an objective to meet " uniformity hypothesis ", i.e.,
If two edges in reference picture are at a distance of closer, the matching degree between edge in this two edges and initial parallax figure
Also should be more similar.With consistent degree matrix X come the consistent degree at the edge in evaluation reference image（Similar to initial parallax figure
The concept of similarity）, X is normalized, C=D ' is obtained-1/2Δ′D′-1/2, wherein Δ '=D '-X, D ' is a diagonal matrix, each
Individual diagonal element is equal to each row element sums of matrix X.
So as to loss equation Ec(F) it is defined as：
Ec(F)=tr{FCFT}
Based on above-mentioned formula, energy equation is defined as follows：
E(F)=tr{(F-Y)T(F-Y)}+αtr{FTLF}+βtr{FCFT}
Target equation is：
F=argmin{E(F)}
Partial differential is asked with F to above formula E (F), and to ask the result after partial differential to be 0（That is, E (F) takes most when what value F take
It is small）, it can obtain,
(αL+I)F+βFC=Y
Above-mentioned formula substantially belongs to Sylvester equations（Reference can be made to webpagehttp://en.wikipedia.org/ wiki/Sylvester_equation）, the equation commonly uses in control theory back warp, will not be described here.If P=α L+1.Wherein,
It is 1 on diagonal that I, which is, and other are 0 diagonal matrix.It is well known that above-mentioned and if only if P characteristic value η1,η2,…ηmWith C's
Characteristic value ζ1,ζ2,…ζn, meet ηi+ζj≠0(i=1,2,…,m;J=1,2 ..., when n), there is unique solution.And this condition is in reality
It is easily met among border.
To matrix F vectorization（The specific steps of vectorization can be found in webpagehttp://en.wikipedia.org/wiki/ Vectorization_(mathematics)）, above-mentioned formula can change into linear equation, as follows：
HereIt is Kronecker multiplication operators, ImAnd InIt is n × n and m × m unit matrix respectively.Vec () is
The vectorization operator of matrix.Finally, we can solve F from vec (F), as follows：
Plus sige herein is inverse ()+It is called group inverse matrices.Traditional matrix inversion, it is necessary to meet two conditions：1. square
Battle array is square formation；2. matrix as matrix's reversibility can just invert.When being unsatisfactory for the two conditions, matrix is just no inverse, but
It is to have so-called " generalized inverse ".It can be regarded as the popularization of inverse matrix, it has many and similar property of inverse matrix.
The strict difinition of generalized inverse is：For matrix A, there is matrix G so that G meets following 4 equations：AGA=A;GAG=G;(AG)'
=AG;(GA) '=GA wherein subscripts ' represent conjugate transposition, then claim the generalized inverse matrix that G is A.It is also known in this field that the plus sige is inverse
, do not describe excessively herein.
In this way, can be in the hope of amended matching degree matrix F.
Finally, based on above-mentioned amended matching degree matrix F(m×n),（A side in initial parallax figure is represented per a line
The matching degree of the difference at each edge of edge and reference picture）Find out in the matrix per a line（Each side of initial parallax figure
Edge）Maximum matching degree（Score）, with maximum matching degree（Score）The edge of corresponding reference picture, is with reference to figure
The edge of matching as in, each edge of initial parallax figure according to the edge of matching obtained above, be modified to and referred to
The edge of matching in image is consistent, to correct initial parallax figure.That is, original each edge according to initial parallax figure is passed through
The parallax value of both sides, is modified to the parallax value for the new both sides of edges that matching is obtained（So that its is consistent）.In this manner it is possible to obtain school
Disparity map after just.
Certainly, above-mentioned formula is only example, rather than limitation.Those skilled in the art can be according to the principle of the present invention come structure
Think new formula, these new formula are also fallen within the scope of the present invention.
Fig. 7 shows the block diagram of the system 700 of correction disparity map according to another embodiment of the present invention.
The system 700 of the correction disparity map includes：Detection means 701, be configured as detecting one of the first disparity map or
One or more second edges of multiple first edges and reference picture；And means for correcting 702, it is configured as being based on being detected
One or more first edges and one or more second edges matching, first disparity map is corrected to obtain after correction
The second disparity map.
In one embodiment, at least one in one or more of first edges and one or more second edges
Length can be more than predetermined length threshold value.
In one embodiment, the means for correcting 702 can include：Based on the one or more first edges detected
With the size and location of one or more second edges, the dress of each first edge and the matching degree of each second edge is determined
Put；Maximum one in matching degree based on specific first edge and all second edges, it is determined that with each first edge
The device of the corresponding second edge of matching；Each described first edge is modified to the dress of its corresponding second edge unanimously
Put；And based on each revised first edge, to correct the device of first disparity map.
In one embodiment, it is described based on the one or more first edges and one or more second edges that are detected
Size and location, determining the device of each first edge and the matching degree of each second edge can be configured as：For
Each specific first edge vi：Select and the specific first edge v in a reference imageiThere is coincidence in a first direction
One or more specific second edges；Calculate the specific first edge viIt is one or more specific with the selection
Second edge the distance in the second direction different from first direction, as the specific first edge respectively with institute
State each preliminary matches degree of one or more specific second edges；And set up preliminary matches degree matrix Y(m×n){yij, this
In m and n be respectively the first edge and second edge number, wherein the preliminary matches degree matrix Y(m×n){yijIn
Element yijIt is the preliminary matches degree of j-th of second edge in i-th of first edge and reference picture in the first disparity map, wherein, i
It is positive integer, and i<=m, and j is positive integer, and j<=n.
In one embodiment, it is described based on the one or more first edges and one or more second edges that are detected
Size and location, determine that the device of each first edge and the matching degree of each second edge is additionally configured to：Based on institute
State in the distribution character of first edge and the distribution character of the second edge in the reference picture in the first disparity map extremely
It is few one, the preliminary matches degree matrix is changed, to obtain amended matching degree.
In one embodiment, the distribution character of the first edge in first disparity map can be obtained by following steps
Arrive：For each two first edge vi′And vj′, calculate their central point cpi′And cpj′Minimum range in a first direction；
If the minimum range on the first direction is less than a distance threshold, central point cp is calculatedi′And cpj′Second direction away from
From being used as described two first edge vi′And vj′Similarity；Set up similarity matrix W(m×m){wi′j′Be used to change described first
Matching degree matrix is walked, m is the number at the edge in the first disparity map, element w therein herei′j′It is the i-th ' in the first disparity map
The similarity degree at individual edge and jth ' individual edge, wherein i ', j ' are positive integers, and i '<=m and j '<=m.
In one embodiment, the distribution character of the second edge in the reference picture can be obtained by following steps
Arrive：For each two second edge ui′′And uj′′, calculate their central point cpi′′And cpj′′Most narrow spacing in a first direction
From；If the minimum range on the first direction is less than a distance threshold, central point cp is calculatedi′′And cpj′′In second direction
Distance, be used as described two second edge ui′′And uj′′Consistent degree；Set up consistent degree matrix X(n×n){xi′′j′′Be used to repair
Change the preliminary matches degree matrix, n is the number at the edge in reference picture, element x therein herei‘’j‘’It is reference picture
In the i-th ' ' individual edge and jth ' ' individual edge the degree of consistency, wherein i ' ', j ' ' be positive integer, and i ' '<=n and j ' '<=
n。
In one embodiment, the distribution character of the first edge based in first disparity map and the reference
At least one in the distribution character of second edge in image, changes the preliminary matches degree matrix, amended to obtain
The device of matching degree is additionally configured to：Based on the first edge in preliminary matches degree matrix and first disparity map
At least one in distribution character and the distribution character of the second edge in the reference picture, builds energy equation；Minimize
The energy equation, draws the amended matching degree at each edge in the first disparity map；Build revised matching degree
Matrix F(m×n), every a line of the matrix is the modification at each edge of an edge in the first disparity map and reference picture
Matching degree afterwards.
In one embodiment, the distribution character of the first edge in based on first disparity map and described with reference to figure
Both distribution characters of second edge as in, change the preliminary matches degree matrix, to obtain amended matching degree
In the case of, the energy equation can be built by equation below：
E(F)=El(F)+αEs(F)+βEc(F),
Wherein, El(F)=tr{(F-Y)T(F-Y) }, Es(F)=tr{FTLF }, Ec(F)=tr{FCFT, α and β are constant, its
In, L=D-1/2ΔD1-/2, wherein Δ=D-W, D is a diagonal matrix, each diagonal element be equal to each row element of matrix W it
With；Wherein, C=D '-1/2Δ′D′-1/2, wherein Δ '=D '-X, D ' is a diagonal matrix, and each diagonal element is equal to matrix X
Each row element sum.
In above formula, part matrix F, Y, C, I, P, L, X, W of omission etc. subscript, but can be with by context
Draw the dimension of these matrixes.
The system 700 can also include being configured as performing each of above-mentioned each method according to an embodiment of the invention
Device, will not be repeated here.
It should be noted that a series of foregoing processing can be performed by hardware or software.Foregoing one is being performed by software
In the case of a little processing, the program for constituting this software is installed from network or recording medium.
The method and system for being used to calculate match measure according to embodiments of the present invention is described by reference to accompanying drawing above.
It should be noted that above-described embodiment is only exemplary, and it is and nonrestrictive.Those skilled in the art completely can be to this
Inventive embodiments carry out various modifications and replacement, without departing from the scope of the present invention.
Claims (9)
1. a kind of method for correcting disparity map, including：
Detecting step, detects one or more first edges of the first disparity map and one or more second sides of reference picture
Edge；And
Aligning step, point of distribution character and one or more second edges based on the one or more first edges detected
The matching of cloth characteristic, corrects first disparity map with the second disparity map after being corrected,
Wherein, the aligning step includes：
Based on the one or more first edges detected and the size and location of one or more second edges, determine each
One edge and the matching degree of each second edge；
Maximum one in matching degree based on specific first edge and all second edges, it is determined that with each first edge
The corresponding second edge of matching；
Each described first edge is modified to its corresponding second edge consistent；And
Based on each revised first edge, to correct first disparity map.
2. method as claimed in claim 1, wherein, in one or more of first edges and one or more second edges extremely
Few one length is more than predetermined length threshold value.
3. method as claimed in claim 1, described based on the one or more first edges detected and one or more second sides
The step of size and location of edge, matching degree for determining each first edge and each second edge, includes：
For each specific first edge vi：
Select and the specific first edge v in a reference imageiThere is the one or more specific of coincidence in a first direction
Second edge；
Calculate the specific first edge viWith one or more specific second edges of the selection with first direction
Distance in different second directions, as the specific first edge respectively with one or more of specific second sides
Each preliminary matches degree of edge；And
Set up preliminary matches degree matrix Y(m×n){yij, m and n are the number of the first edge and second edge respectively here, its
Described in preliminary matches degree matrix Y(m×n){yijIn element yijIt is i-th of first edge and reference picture in the first disparity map
In j-th of second edge preliminary matches degree, wherein, i is positive integer, and i<=m, and j is positive integer, and j<=n.
4. method as claimed in claim 3, wherein, it is described based on the one or more first edges detected and one or more
The size and location at two edges, the step of determining the matching degree of each first edge and each second edge also includes：
The distribution of second edge in distribution character and the reference picture based on the first edge in first disparity map
At least one in characteristic, changes the preliminary matches degree matrix, to obtain amended matching degree.
5. method as claimed in claim 4, wherein, the distribution character of the first edge in first disparity map passes through following steps
Obtain：
For each two first edge vi′And vj′, calculate their central point cpi′And cpj′Minimum range in a first direction；
If the minimum range on the first direction is less than a distance threshold, central point cp is calculatedi′And cpj′In second direction
Distance, is used as described two first edge vi′And vj′Similarity；
Set up similarity matrix W(m×m){wi′j′Be used to change the preliminary matches degree matrix, during m is the first disparity map here
The number at edge, element w thereini′j′It is the similarity degree at the i-th ' individual edge and jth ' individual edge in the first disparity map, wherein
I ', j ' are positive integers, and i '<=m and j '<=m.
6. method as claimed in claim 4, wherein, the distribution character of the second edge in the reference picture is obtained by following steps
Arrive：
For each two second edge ui″And uj″, calculate their central point cpi″And cpj″Minimum range in a first direction；
If the minimum range on the first direction is less than a distance threshold, central point cp is calculatedi″And cpj″In second direction
Distance, is used as described two second edge ui″And uj″Consistent degree；
Set up consistent degree matrix X(n×n){xi″j″Be used to change the preliminary matches degree matrix, n is the side in reference picture here
The number of edge, element x thereini‘’j‘’It is the degree of consistency at i-th " individual edge and jth " individual edge in reference picture, wherein
I ", j " is positive integer, and i "<=n and j "<=n.
7. method as claimed in claim 4, the distribution character and the ginseng of the first edge based in first disparity map
At least one in the distribution character of the second edge in image is examined, the preliminary matches degree matrix is changed, to obtain after modification
Matching degree the step of also include：
Distribution character and the reference picture based on the first edge in preliminary matches degree matrix and first disparity map
In second edge distribution character at least one, build energy equation；
The energy equation is minimized, the amended matching degree at each edge in the first disparity map is drawn；
Build revised matching degree matrix F(m×n), every a line of the matrix be an edge in the first disparity map with reference to figure
The amended matching degree at each edge of picture.
8. method as claimed in claim 7, the distribution character of the first edge in based on first disparity map and the reference
Both distribution characters of second edge in image, change the preliminary matches degree matrix, to obtain amended matching degree
In the case of, the energy equation is built by equation below：
E (F)=El(F)+αEs(F)+βEc(F),
Wherein, El(F)=tr F-Y)T(F-Y) }, Es(F)=tr { FTLF }, Ec(F)=tr { FCFT, α and β are constant,
Wherein, L=D-1/2ΔD-1/2, wherein Δ=D-W, D is a diagonal matrix, and it is each that each diagonal element is equal to matrix W
Row element sum；Wherein, C=D '-1/2Δ′D′-1/2, wherein Δ '=D '-X, D ' is a diagonal matrix, each diagonal element
Equal to each row element sums of matrix X, wherein matrix W represents the similarity matrix of the first disparity map, and matrix X is reference picture
Consistent degree matrix.
9. a kind of system for correcting disparity map, including
Detection means, be configured as detect the first disparity map one or more first edges and reference picture it is one or more
Second edge；And
Means for correcting, is configured as the distribution character and one or more second of one or more first edges based on being detected
The matching of the distribution character at edge, corrects first disparity map with the second disparity map after being corrected,
Wherein, the means for correcting includes：
Based on the one or more first edges detected and the size and location of one or more second edges, determine each
One edge and the device of the matching degree of each second edge；
Maximum one in matching degree based on specific first edge and all second edges, it is determined that with each first edge
The device of the corresponding second edge of matching；
Each described first edge is modified to the consistent device of its corresponding second edge；And based on it is revised each
First edge, to correct the device of first disparity map.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310269280.1A CN104252701B (en) | 2013-06-28 | 2013-06-28 | Correct the method and system of disparity map |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310269280.1A CN104252701B (en) | 2013-06-28 | 2013-06-28 | Correct the method and system of disparity map |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104252701A CN104252701A (en) | 2014-12-31 |
| CN104252701B true CN104252701B (en) | 2017-08-29 |
Family
ID=52187569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310269280.1A Expired - Fee Related CN104252701B (en) | 2013-06-28 | 2013-06-28 | Correct the method and system of disparity map |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104252701B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106157285B (en) * | 2015-04-03 | 2018-12-21 | 株式会社理光 | For selecting the method and system of the preferred value of the parameter group for disparity computation |
| CN108205810B (en) * | 2016-12-16 | 2021-08-10 | 富士通株式会社 | Image comparison device and method and electronic equipment |
| CN110223257B (en) * | 2019-06-11 | 2021-07-09 | 北京迈格威科技有限公司 | Method and device for acquiring disparity map, computer equipment and storage medium |
| CN112116660B (en) * | 2019-06-19 | 2024-03-29 | 京东方科技集团股份有限公司 | Disparity map correction method, device, terminal and computer readable medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102999913A (en) * | 2012-11-29 | 2013-03-27 | 清华大学深圳研究生院 | Local three-dimensional matching method based on credible point spreading |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100899422B1 (en) * | 2007-08-06 | 2009-05-27 | 삼성모바일디스플레이주식회사 | The stereo matching system and the stereo matching method of the same |
-
2013
- 2013-06-28 CN CN201310269280.1A patent/CN104252701B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102999913A (en) * | 2012-11-29 | 2013-03-27 | 清华大学深圳研究生院 | Local three-dimensional matching method based on credible point spreading |
Non-Patent Citations (3)
| Title |
|---|
| LIU Zhengdong et al..Edge Traction Stereo Matching Method Using RBF.《2006 8th International Conference on Signal Processing》.2006,第2卷第1-4页. * |
| 一种快速有效提取浓密视差图的方法;廖素英 等;《电子科技大学学报》;20050430;第34卷(第2期);第198-201页 * |
| 侯建 等.月球车视觉系统立体匹配算法研究.《南京理工大学学报（自然科学版）》.2008,第32卷(第2期),第176-180页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104252701A (en) | 2014-12-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11900628B2 (en) | Stereo matching method and apparatus, image processing apparatus, and training method therefor | |
| CN105528785B (en) | A kind of binocular vision image solid matching method | |
| US10929998B2 (en) | Method and apparatus for estimating disparity | |
| CN108876749A (en) | A kind of lens distortion calibration method of robust | |
| EP2597597A2 (en) | Apparatus and method for calculating three dimensional (3D) positions of feature points | |
| CN106887018B (en) | Stereo matching method, controller and system | |
| CN108537837A (en) | A kind of method and relevant apparatus of depth information determination | |
| CN102855459B (en) | For the method and system of the detection validation of particular prospect object | |
| CN103310421B (en) | The quick stereo matching process right for high-definition image and disparity map acquisition methods | |
| CN105513064A (en) | Image segmentation and adaptive weighting-based stereo matching method | |
| CN110909693A (en) | 3D face living body detection method and device, computer equipment and storage medium | |
| CN104252701B (en) | Correct the method and system of disparity map | |
| CN103996202A (en) | Stereo matching method based on hybrid matching cost and adaptive window | |
| CN104021559A (en) | Image registration method based on mutual information and Harris corner point detection | |
| CN104318576A (en) | Super-pixel-level image global matching method | |
| CN108470356A (en) | A kind of target object fast ranging method based on binocular vision | |
| CN103458261A (en) | Video scene variation detection method based on stereoscopic vision | |
| CN102740096A (en) | Space-time combination based dynamic scene stereo video matching method | |
| CN105761270A (en) | Tree type filtering three-dimensional coupling method based on epipolar line linear distance transformation | |
| CN112365586A (en) | 3D face modeling and stereo judging method and binocular 3D face modeling and stereo judging method of embedded platform | |
| CN104200453A (en) | Parallax image correcting method based on image segmentation and credibility | |
| CN113313740B (en) | Disparity map and surface normal vector joint learning method based on plane continuity | |
| CN105138979A (en) | Method for detecting the head of moving human body based on stereo visual sense | |
| CN116229123A (en) | Binocular stereo matching method and device based on multi-channel grouping cross-correlation cost volume | |
| EP3076370B1 (en) | Method and system for selecting optimum values for parameter set for disparity calculation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170829 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |