CN107730543A - A kind of iteratively faster computational methods of half dense stereo matching - Google Patents
A kind of iteratively faster computational methods of half dense stereo matching Download PDFInfo
- Publication number
- CN107730543A CN107730543A CN201710805755.2A CN201710805755A CN107730543A CN 107730543 A CN107730543 A CN 107730543A CN 201710805755 A CN201710805755 A CN 201710805755A CN 107730543 A CN107730543 A CN 107730543A
- Authority
- CN
- China
- Prior art keywords
- point
- mrow
- image
- left image
- matching
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Length Measuring Devices By Optical Means (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to technical field of computer vision, and it discloses a kind of iteratively faster computational methods of half dense stereo matching, while meeting to obtain compared with high matching precision situation, improve matching speed.This method can be summarized as:Left image, the characteristic point of right image, construction feature description are extracted respectively;Then, the strong point is referred to as according to Feature Descriptor and epipolar-line constraint, the Feature Points Matching of completion left and right figure, the characteristic point that the match is successful;Then, Delaunay triangles are built according to the strong point in left image, estimates the priori parallax d of left image all pixels pointi;And calculate priori parallax diCorresponding cost error Ci, obtain the minimum cost error C of all strong points of current iterationmin;To support point set and the continuous iteration renewal of minimum cost error, until meeting iteration stopping condition, finally, according to its match point in right image of the disparity computation of each pixel of left image.
Description
Technical field
The present invention relates to technical field of computer vision, and in particular to a kind of iteratively faster of half dense stereo matching calculates
Method.
Background technology
Image Stereo matching is computer vision, photogrammetric, computer graphics subject a important branch, is being permitted
There is highly important value in more applications.Images match can be divided into sparse matching (sparse matching) and dense Stereo Matching
(dense matching).Sparse matching is usually the characteristic point for having on extraction image stronger texture, is then retouched by feature
Sub- calculating Matching power flow is stated, obtains optimal matching.Openness due to characteristic point, sparse matching can not carry in many applications
For the characteristic point and three-dimensional point of enough numbers, therefore relatively limited three-dimensional world information can only be obtained.Dense Stereo Matching is to figure
Each pixel of picture is matched, therefore can obtain intensive three-dimensional world information.The algorithm of dense Stereo Matching can be divided into entirely
Office's method and the class of partial approach two.In recent years, as the new algorithm of partial approach continues to bring out and improves, performance is constantly lifted,
The method of dense Stereo Matching has been able to (be applied to work as after e.g., some dense Stereo Matching algorithms are optimized applied to some occasions in real time
On preceding mobile process chip (such as ARM and Movidius), the treatment effeciency of QVGA images can reach 30fps's).It is but intensive
Match due to its intrinsic calculating limitation, be unable to reach higher processing frame frequency and (VGA figures e.g., obtained under the conditions of identity of operation
As efficiency 30fps or higher).
Therefore, there is following defect in the above two stereoscopic scheme in conventional art:Sparse matching treatment efficiency is higher, but
Sufficient amount of match point information can not be provided;Dense Stereo Matching can obtain intensive match point information, but can not obtain higher
Treatment effeciency, more and more high industry requirement can not be adapted to.
The content of the invention
The technical problems to be solved by the invention are:A kind of iteratively faster computational methods of half dense stereo matching are proposed,
While meeting to obtain compared with high matching precision situation, matching speed is improved.
Scheme is used by the present invention solves above-mentioned technical problem:
A kind of iteratively faster computational methods of half dense stereo matching, comprise the following steps:
A. left image and right image are obtained using binocular camera;
B. left image, the characteristic point of right image, construction feature description are extracted respectively;
C. according to Feature Descriptor and epipolar-line constraint, the Feature Points Matching of completion left and right figure, the characteristic point that the match is successful
The referred to as strong point;
D. Delaunay triangles are built according to the strong point in left image;
E. the priori parallax d of left image all pixels point is estimatedi;
F. the priori parallax d of each pixel is calculatediCorresponding cost error Ci, obtain all strong points of current iteration
Minimum cost error Cmin;
G. it is control point to obtain reliable match point, and multiple control points that reliability arrangement ranks forefront are arranged into new support
Point, renewal support point set and minimum cost error Cmin;
H. judge whether to meet iteration stopping condition, if satisfied, then stopping iterative calculation, into step i;If not satisfied,
Then return to step d, continue to iterate to calculate;
I. according to its match point in right image of the disparity computation of each pixel of left image.
As further optimization, in step a, image that the left camera of binocular camera obtains is left image, right shooting
The image that head obtains is right image.
Optimize as further, in step b, characteristic point is carried out to left image and right image respectively using SIFT algorithms and carried
Take, specifically include:
B1. multiple secondary continuous filterings are carried out to left image using Gaussian filter and establishes first yardstick group;
B2. left image is reduced to original half and carries out same gaussian filtering second yardstick group of formation;
B3. left image is reduced to the same gaussian filtering of 1/4 original progress and forms the 3rd yardstick group;
B4. by that analogy, untill left image is less than a certain given threshold;
B5. difference, Gaussian difference scale group corresponding to formation are carried out to the gaussian filtering image in each yardstick group;
B6. the local extremum of the image in Gaussian difference scale group is taken, obtains the characteristic point of left image on metric space domain,
Characteristic point is stored in characteristic vector in order, obtains left image Feature Descriptor;
B7. mode of the right image with reference to step b1~b6 is operated, obtains the characteristic point of right image on metric space domain, will
Characteristic point is stored in characteristic vector in order, obtains right image Feature Descriptor.
As further optimization, step c is specifically included:
C1. left image and right image are corrected, are met the image characteristic point of epipolar-line constraint;
C2. Feature Descriptor matches:Find out and characteristic point p in left imageiEuclidean distance is recently and time near in right image
In two adjacent features point descriptor q 'iWith q "i, calculate q 'iWith piAnd q "iWith piThe ratio of Euclidean distance between two groups of characteristic points
Value r, this group of characteristic point is considered as that the match is successful if ratio r is less than defined threshold, and this group of characteristic point otherwise is considered as into matching loses
Lose, the point that it fails to match is no longer characteristic point;Using two characteristic points that the match is successful as sparse matching same place pair, respectively will
Two characteristic points of sparse matching same place pair are as the first sparse matching same place and the second sparse matching same place.
Optimize as further, it is described that Delaunay triangles are built according to the strong point in left image in step d, specifically
Including:
Using the first sparse matching same place as the strong point, Delaunay Triangulation is carried out to left image, obtained multiple
Delaunay triangles.
Optimize as further, in step e, for the parallax of arbitrfary point p in Delaunay triangles, by the coordinate of the point
The fitting parameter that Delaunay triangle projective planums are sitting in the point is estimated to obtain:
dp=aup+bvp+c
Wherein, parameter a, b, c are obtained by Delaunay triangle projective planums where being fitted P points, (up,vp) for P points in image
In coordinate.
Optimize as further, in step f, calculate the priori parallax d of each pixeliCorresponding cost error Ci, obtain
The minimum cost error C of all strong points of current iterationmin, specifically include:
The sampling window that size is N*N is chosen, calculates each pixel i priori parallax diCorresponding cost error Ci:
Wherein, Ij,IiThe respectively pixel value of point i pixel value and point j, point j is in the N*N windows centered on point i;
Then, the minimum cost error C of all strong points of current iteration is obtainedmin:
Optimize as further, in step g, the more reliable match point of acquisition is for the method at control point:
Meet for given match error threshold l, such as fruit dot i cost error:Ci< l | | Ci< Cmin, then it is by the point
Reliably, reliability is arranged into forward multiple control points and includes support point set;
The renewal minimum cost error CminMode be:
Wherein, I-i represents that left image I removes the set of all pixels point beyond point i.
Optimize as further, in step h, the iteration stopping condition is:Support the quantity of the strong point in point set
Reach given threshold M.
The beneficial effects of the invention are as follows:
Support point set as dense as possible is obtained by continuous iteration, renewal, can be based on Delaunay triangles
It is more accurate for the disparity estimation of non-point of interest, so as to improve images match precision in the case where ensureing rate matched,
High-precision disparity map, the application field high especially suitable for mobile platform or requirement of real-time can quickly be obtained.
Brief description of the drawings
Fig. 1 is the iteratively faster computational methods flow chart of the half dense stereo matching in the embodiment of the present invention.
Embodiment
The present invention is directed to propose a kind of iteratively faster computational methods of half dense stereo matching, are meeting to obtain higher matching
While precise manner, matching speed is improved.
Below in conjunction with the accompanying drawings and embodiment the solution of the present invention is further described:
Embodiment:
Walked as shown in figure 1, the iteratively faster computational methods of the half dense stereo matching in the present embodiment include following implementation
Suddenly:
S1:Left image and right image are obtained using binocular camera;
The image that the left camera of binocular camera obtains is referred to as left image in the present embodiment, by the right side of binocular camera
The image that camera obtains is referred to as right image.
S2:Left image, the characteristic point of right image, construction feature description are extracted respectively;
The present embodiment is carried out special to left image and right image respectively using SIFT algorithms (Scale invariant features transform algorithm)
Sign point extraction:
Multiple secondary continuous filterings first are carried out to left image or right image using Gaussian filter and establish first yardstick group;
Left image or right image are reduced to original half again and carry out same gaussian filtering second yardstick group of formation;Again left figure
Picture or right image are reduced to the same gaussian filtering of 1/4 original progress and form the 3rd yardstick group, repeat by that analogy
Untill left image or right image are less than a certain given threshold value;Next difference is carried out to the Gaussian image in each yardstick group
Form Gaussian difference scale group.Then the local extremum in these difference of Gaussian left images or right image is taken just to obtain yardstick sky
Between the characteristic point of left image or right image on domain.When all carrying out above operation to left image and right image, you can obtain left image
Characteristic point and right image characteristic point.Characteristic point is stored in characteristic vector in order, obtains Feature Descriptor.
S3:According to Feature Descriptor and epipolar-line constraint, the Feature Points Matching of completion left and right figure;The characteristic point that the match is successful
The referred to as strong point;
This step when it is implemented, including:
S301:Left image and right image are corrected, are met the image characteristic point of epipolar-line constraint;
S302:Feature Descriptor matches:Feature Descriptor to left image and carried out between the Feature Descriptor of right image
Matching, the present embodiment are realized between the Feature Descriptor of left image and the Feature Descriptor of right image using SIFT matching algorithms
Matching:Find out and characteristic point p in left imageiEuclidean distance is retouched with time near two adjacent features points in right image recently
State symbol q 'iWith q "i, calculate q 'iWith piAnd q "iWith piThe ratio r of Euclidean distance between two groups of characteristic points, if ratio r is less than
This group of characteristic point is then considered as that the match is successful by defined threshold, otherwise this group of characteristic point is considered as to the point that it fails to match, and it fails to match
No longer it is characteristic point.Using two characteristic points that the match is successful as sparse matching same place pair, respectively by sparse matching same place
To two characteristic points as the first sparse matching same place and the second sparse matching same place.
S4:In left image, Delaunay triangles are built according to the strong point;
In this step, using the first sparse matching same place as the strong point, Delaunay Triangulation is carried out to left image, obtained
To multiple Delaunay triangles.
S5:Utilize the priori parallax of left image Delaunay triangles estimation left image all pixels point.
In this step, for the parallax of any point p in Delaunay triangles, it is sitting in by the coordinate and the point of the point
The fitting parameter of Delaunay triangle projective planums is estimated to obtain, shown in mathematic(al) representation specific as follows:
dp=aup+bvp+c
Wherein, parameter a, b, c are obtained by Delaunay triangle projective planums where being fitted the point, (up,vp) scheming for the point
Coordinate as in.
S6:Calculate the priori parallax d of each pixeliCorresponding cost error Ci, obtain all strong points of current iteration
Minimum cost error Cmin。
The present embodiment carries out characteristic point local matching using CT algorithms (Census Transform), specifically, choosing big
The small sampling window for N*N, calculate the priori parallax d of each pixel i (including strong point)iCorresponding cost error Ci, obtain
The minimum cost error C of all strong points of current iterationmin, specifically:
Wherein, Ij,IiThe respectively pixel value of point i pixel value and point j, point j is in the N*N windows centered on point i.
S7:It is control point to obtain more reliable match point, and multiple control points that reliability arrangement ranks forefront are arranged into new
The strong point, renewal support point set and minimum cost error Cmin;
This step in the specific implementation, gives match error threshold l, as fruit dot i cost error meets:Ci< l | | Ci<
Cmin, then it is reliable by the point, reliability is arranged into forward multiple control points includes support point set;Meanwhile in left image
Minimum cost error C is updated in Imin:
Wherein, I-i represents that left image I removes the set of all pixels point beyond point i.
S8:Judge whether to meet iteration stopping condition, if satisfied, then stopping iterative calculation, into step S9;It is if discontented
Foot, then return to step S4, continues to iterate to calculate;
This step in the specific implementation, sets a threshold value M, supports the number of support points in point set to reach given threshold
M, then stop calculating, into step S9;Otherwise, S4 is returned, continues to iterate to calculate.
S9:According to its match point in right image of the disparity computation of each pixel of left image.
It is prior art with specific reference to disparity computation match point, the present embodiment is not repeated this.
Support point set as dense as possible is obtained by continuous iteration, renewal in above-described embodiment scheme, can be caused
It is more accurate for the disparity estimation of non-point of interest based on Delaunay triangles, so as to be lifted in the case where ensureing rate matched
Images match precision, can quickly obtain high-precision disparity map.
Claims (9)
1. a kind of iteratively faster computational methods of half dense stereo matching, it is characterised in that comprise the following steps:
A. left image and right image are obtained using binocular camera;
B. left image, the characteristic point of right image, construction feature description are extracted respectively;
C. it is referred to as according to Feature Descriptor and epipolar-line constraint, the Feature Points Matching of completion left and right figure, the characteristic point that the match is successful
The strong point;
D. Delaunay triangles are built according to the strong point in left image;
E. the priori parallax d of left image all pixels point is estimatedi;
F. the priori parallax d of each pixel is calculatediCorresponding cost error Ci, obtain all strong points of current iteration most
Small cost error Cmin;
G. it is control point to obtain reliable match point, and multiple control points that reliability arrangement ranks forefront are arranged into the new strong point,
Renewal support point set and minimum cost error Cmin;
H. judge whether to meet iteration stopping condition, if satisfied, then stopping iterative calculation, into step i;If not satisfied, then return
Step d is returned, continues to iterate to calculate;
I. according to its match point in right image of the disparity computation of each pixel of left image.
A kind of 2. iteratively faster computational methods of half dense stereo matching as claimed in claim 1, it is characterised in that step a
In, the image that the left camera of binocular camera obtains is left image, and the image that right camera obtains is right image.
A kind of 3. iteratively faster computational methods of half dense stereo matching as claimed in claim 1, it is characterised in that step b
In, feature point extraction is carried out to left image and right image respectively using SIFT algorithms, specifically included:
B1. multiple secondary continuous filterings are carried out to left image using Gaussian filter and establishes first yardstick group;
B2. left image is reduced to original half and carries out same gaussian filtering second yardstick group of formation;
B3. left image is reduced to the same gaussian filtering of 1/4 original progress and forms the 3rd yardstick group;
B4. by that analogy, untill left image is less than a certain given threshold;
B5. difference, Gaussian difference scale group corresponding to formation are carried out to the gaussian filtering image in each yardstick group;
B6. the local extremum of the image in Gaussian difference scale group is taken, the characteristic point of left image on metric space domain is obtained, by spy
Sign point is stored in characteristic vector in order, obtains left image Feature Descriptor;
B7. mode of the right image with reference to step b1~b6 is operated, the characteristic point of right image on metric space domain is obtained, by feature
Point is stored in characteristic vector in order, obtains right image Feature Descriptor.
A kind of 4. iteratively faster computational methods of half dense stereo matching as claimed in claim 1, it is characterised in that step c
Specifically include:
C1. left image and right image are corrected, are met the image characteristic point of epipolar-line constraint;
C2. Feature Descriptor matches:Find out and characteristic point p in left imageiEuclidean distance is recently and time near two in right image
Adjacent features point descriptor q 'iWith q "i, calculate q 'iWith piAnd q "iWith piThe ratio r of Euclidean distance between two groups of characteristic points, such as
Fruit ratio r is less than defined threshold, and then this group of characteristic point is considered as that the match is successful, otherwise this group of characteristic point be considered as to it fails to match, matching
The point of failure is no longer characteristic point;Using two characteristic points that the match is successful as sparse matching same place pair, respectively by sparse
Two characteristic points with same place pair are as the first sparse matching same place and the second sparse matching same place.
A kind of 5. iteratively faster computational methods of half dense stereo matching as claimed in claim 1, it is characterised in that step d
In, it is described that Delaunay triangles are built according to the strong point in left image, specifically include:
Using the first sparse matching same place as the strong point, Delaunay Triangulation is carried out to left image, obtained multiple
Delaunay triangles.
A kind of 6. iteratively faster computational methods of half dense stereo matching as claimed in claim 5, it is characterised in that step e
In, for the parallax of arbitrfary point p in Delaunay triangles, Delaunay triangle projective planums are sitting in by the coordinate and the point of the point
Fitting parameter estimation obtain:
dp=aup+bvp+c
Wherein, parameter a, b, c are obtained by Delaunay triangle projective planums where being fitted P points, (up,vp) for P points in the picture
Coordinate.
A kind of 7. iteratively faster computational methods of half dense stereo matching as claimed in claim 6, it is characterised in that step f
In, calculate the priori parallax d of each pixeliCorresponding cost error Ci, obtain minimum generations of all strong points of current iteration
Valency error Cmin, specifically include:
The sampling window that size is N*N is chosen, calculates each pixel i priori parallax diCorresponding cost error Ci:
<mrow>
<msub>
<mi>C</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>*</mo>
<mi>N</mi>
</mrow>
</munderover>
<mrow>
<mo>|</mo>
<mrow>
<msub>
<mi>I</mi>
<mi>j</mi>
</msub>
<mo>-</mo>
<msub>
<mi>I</mi>
<mi>i</mi>
</msub>
</mrow>
<mo>|</mo>
</mrow>
</mrow>
Wherein, Ij,IiThe respectively pixel value of point i pixel value and point j, point j is in the N*N windows centered on point i;
Then, the minimum cost error C of all strong points of current iteration is obtainedmin:
<mrow>
<msub>
<mi>C</mi>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mrow>
<mi>i</mi>
<mo>&Element;</mo>
<mi>I</mi>
</mrow>
</munder>
<mrow>
<mo>(</mo>
<msub>
<mi>C</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
A kind of 8. iteratively faster computational methods of half dense stereo matching as claimed in claim 7, it is characterised in that step g
In, the more reliable match point of acquisition is for the method at control point:
Meet for given match error threshold l, such as fruit dot i cost error:Ci< l | | Ci< Cmin, then it is reliable by the point
, reliability is arranged into forward multiple control points and includes support point set;
The renewal minimum cost error CminMode be:
<mrow>
<msub>
<mi>C</mi>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mrow>
<mi>k</mi>
<mo>&Element;</mo>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>-</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
</mrow>
</munder>
<mrow>
<mo>(</mo>
<msub>
<mi>C</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Wherein, I-i represents that left image I removes the set of all pixels point beyond point i.
A kind of 9. iteratively faster computational methods of half dense stereo matching as claimed in claim 8, it is characterised in that step h
In, the iteration stopping condition is:The quantity of the strong point in support point set reaches given threshold M.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710805755.2A CN107730543B (en) | 2017-09-08 | 2017-09-08 | Rapid iterative computation method for semi-dense stereo matching |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710805755.2A CN107730543B (en) | 2017-09-08 | 2017-09-08 | Rapid iterative computation method for semi-dense stereo matching |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107730543A true CN107730543A (en) | 2018-02-23 |
CN107730543B CN107730543B (en) | 2021-05-14 |
Family
ID=61205786
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710805755.2A Active CN107730543B (en) | 2017-09-08 | 2017-09-08 | Rapid iterative computation method for semi-dense stereo matching |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107730543B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109816706A (en) * | 2019-02-01 | 2019-05-28 | 辽宁工程技术大学 | A kind of smoothness constraint and triangulation network equal proportion subdivision picture are to dense matching method |
GB2583839A (en) * | 2018-04-05 | 2020-11-11 | Imagination Tech Ltd | Matching local image feature descriptors |
US10885376B2 (en) | 2018-04-05 | 2021-01-05 | Imagination Technologies Limited | Matching local image feature descriptors in image analysis |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101393497A (en) * | 2008-10-30 | 2009-03-25 | 上海交通大学 | Multi-point touch method based on binocular stereo vision |
CN101720047A (en) * | 2009-11-03 | 2010-06-02 | 上海大学 | Method for acquiring range image by stereo matching of multi-aperture photographing based on color segmentation |
CN103700099A (en) * | 2013-12-18 | 2014-04-02 | 同济大学 | Rotation and dimension unchanged wide baseline stereo matching method |
CN104167000A (en) * | 2014-08-25 | 2014-11-26 | 同济大学 | Affine-invariant wide-baseline image dense matching method |
EP2897101A1 (en) * | 2014-01-20 | 2015-07-22 | Nokia Corporation | Visual perception matching cost on binocular stereo images |
CN106780442A (en) * | 2016-11-30 | 2017-05-31 | 成都通甲优博科技有限责任公司 | A kind of solid matching method and system |
-
2017
- 2017-09-08 CN CN201710805755.2A patent/CN107730543B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101393497A (en) * | 2008-10-30 | 2009-03-25 | 上海交通大学 | Multi-point touch method based on binocular stereo vision |
CN101720047A (en) * | 2009-11-03 | 2010-06-02 | 上海大学 | Method for acquiring range image by stereo matching of multi-aperture photographing based on color segmentation |
CN103700099A (en) * | 2013-12-18 | 2014-04-02 | 同济大学 | Rotation and dimension unchanged wide baseline stereo matching method |
EP2897101A1 (en) * | 2014-01-20 | 2015-07-22 | Nokia Corporation | Visual perception matching cost on binocular stereo images |
CN104167000A (en) * | 2014-08-25 | 2014-11-26 | 同济大学 | Affine-invariant wide-baseline image dense matching method |
CN106780442A (en) * | 2016-11-30 | 2017-05-31 | 成都通甲优博科技有限责任公司 | A kind of solid matching method and system |
Non-Patent Citations (3)
Title |
---|
JIANBO JIAO ET AL.: "cost-volume filtering-based stereo matching with improved matching cost and secondary refinement", 《2014 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO(ICME)》 * |
葛忠孝等: "基于树形结构的半全局立体匹配算法", 《计算机工程》 * |
贾永红: "《数字图像处理》", 31 July 2015, 武汉:武汉大学出版社 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2583839A (en) * | 2018-04-05 | 2020-11-11 | Imagination Tech Ltd | Matching local image feature descriptors |
US10885376B2 (en) | 2018-04-05 | 2021-01-05 | Imagination Technologies Limited | Matching local image feature descriptors in image analysis |
GB2583839B (en) * | 2018-04-05 | 2021-05-05 | Imagination Tech Ltd | Matching local image feature descriptors |
US11756222B2 (en) | 2018-04-05 | 2023-09-12 | Imagination Technologies Limited | Matching local image feature descriptors in image analysis |
CN109816706A (en) * | 2019-02-01 | 2019-05-28 | 辽宁工程技术大学 | A kind of smoothness constraint and triangulation network equal proportion subdivision picture are to dense matching method |
CN109816706B (en) * | 2019-02-01 | 2022-12-06 | 辽宁工程技术大学 | Smooth constraint and triangulation equal-proportion image pair dense matching method |
Also Published As
Publication number | Publication date |
---|---|
CN107730543B (en) | 2021-05-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110220493B (en) | Binocular distance measuring method and device | |
CN108537848B (en) | Two-stage pose optimization estimation method for indoor scene reconstruction | |
CN112991447B (en) | Visual positioning and static map construction method and system in dynamic environment | |
CN103440653A (en) | Binocular vision stereo matching method | |
CN111914913B (en) | Novel stereo matching optimization method | |
CN113362247A (en) | Semantic live-action three-dimensional reconstruction method and system of laser fusion multi-view camera | |
CN109389621B (en) | RGB-D target tracking method based on multi-mode depth feature fusion | |
CN110942476A (en) | Improved three-dimensional point cloud registration method and system based on two-dimensional image guidance and readable storage medium | |
CN112862877A (en) | Method and apparatus for training image processing network and image processing | |
WO2017207647A1 (en) | Apparatus and method for performing 3d estimation based on locally determined 3d information hypotheses | |
CN107730543A (en) | A kind of iteratively faster computational methods of half dense stereo matching | |
EP3293700A1 (en) | 3d reconstruction for vehicle | |
CN106408596A (en) | Edge-based local stereo matching method | |
CN105225233B (en) | A kind of stereopsis dense Stereo Matching method and system based on the expansion of two classes | |
CN110942102B (en) | Probability relaxation epipolar matching method and system | |
CN103337064A (en) | Method for removing mismatching point in image stereo matching | |
Shi et al. | Stereo matching using local plane fitting in confidence-based support window | |
CN112819874B (en) | Depth information processing method, apparatus, device, storage medium, and program product | |
Zhu et al. | PairCon-SLAM: Distributed, online, and real-time RGBD-SLAM in large scenarios | |
KR100792172B1 (en) | Apparatus and method for estimating fundamental matrix using robust correspondence point | |
CN110610503B (en) | Three-dimensional information recovery method for electric knife switch based on three-dimensional matching | |
CN109816710B (en) | Parallax calculation method for binocular vision system with high precision and no smear | |
CN112767481A (en) | High-precision positioning and mapping method based on visual edge features | |
CN109785367B (en) | Method and device for filtering foreign points in three-dimensional model tracking | |
CN116429087A (en) | Visual SLAM method suitable for dynamic environment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |