CN106952304A - A kind of depth image computational methods of utilization video sequence interframe correlation - Google Patents
A kind of depth image computational methods of utilization video sequence interframe correlation Download PDFInfo
- Publication number
- CN106952304A CN106952304A CN201710172103.XA CN201710172103A CN106952304A CN 106952304 A CN106952304 A CN 106952304A CN 201710172103 A CN201710172103 A CN 201710172103A CN 106952304 A CN106952304 A CN 106952304A
- Authority
- CN
- China
- Prior art keywords
- parallax
- point
- frame
- disparity
- value
- 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
- 238000000205 computational method Methods 0.000 title claims abstract description 14
- 238000005457 optimization Methods 0.000 claims abstract description 23
- 238000004364 calculation method Methods 0.000 claims abstract description 13
- 238000000034 method Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 19
- 230000009466 transformation Effects 0.000 claims description 3
- 238000006116 polymerization reaction Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000012805 post-processing Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000001027 hydrothermal synthesis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012372 quality testing Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- 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
- G06T2207/10016—Video; Image sequence
Landscapes
- Measurement Of Optical Distance (AREA)
Abstract
The present invention discloses a kind of depth image computational methods of utilization video sequence interframe correlation, including：Step 1, the picture to the shooting of left and right view camera are corrected；Step 2, for the first frame left and right picture point p, calculate Matching power flow in maximum disparity hunting zone, carry out parallax optimization, parallax refinement, obtain initial parallax value d1；The gradient of disparity of first frame is g1；Step 3, for next frame p points calculating, when p points are not abnormity point or marginal point, disparity search scope is set on the basis of the parallax value of previous frame p points, otherwise Matching power flow is calculated with maximum disparity hunting zone, carry out parallax optimization, parallax refinement, and the parallax value and gradient of disparity of present frame are obtained, realize real-time Stereo matching.The present invention significantly reduces the amount of calculation of present frame algorithm using the parallax of previous frame in video sequence, can reduce and takes under conditions of original algorithmic match precision is ensured, improve the efficiency in practical application scene.
Description
Technical field
The present invention relates to computer stereo vision field, more particularly to a kind of utilize video sequence interframe correlation
Depth image computational methods.
Background technology
Stereoscopic vision is one of widest research topic in computer vision field in recent years.It is from Same Scene not
The technology of depth information of scene is obtained in the picture shot with viewpoint.The picture shot from two cameras of different points of view is obtained,
The horizontal displacement of the corresponding points o'clock on two width pictures in scene is calculated for this two width picture, this process is thus referred to as double
Mesh Stereo matching, horizontal displacement is thus referred to as parallax.Stereoscopic vision is widely used in many fields, for example obstacle quality testing
Survey, drive auxiliary, three-dimensional reconstruction and motion detection etc..
With the development of consumer electronics, the scene of stereoscopic vision application is more and more extensive.As shown in figure 1, in general, greatly
Most Stereo Matching Algorithms may be summarized to be four steps：
(1) Matching power flow is calculated
(2) cost polymerize
(3) parallax optimization/calculating
(4) parallax refinement (post processing)
And the implementation detail of specific steps depends on algorithm in itself.According to step 1,2,3 difference, most of algorithms are divided into
Two kinds of main Types, local algorithm and Global Algorithm.Local algorithm is based primarily upon window and carries out cost polymerization (step 2), for
Obtained energy value is calculated in one given point, parallax optimization (step 3) and is dependent only on owning in a restriction window
The pixel value of point.Local algorithm is divided into step 1,2,3 with being generally apparent from.For example, traditional difference of two squares summation algorithm (sum-
Of-squared-differences), its calculation procedure is as follows：
(1) Matching power flow is to calculate corresponding in the figure of left side viewpoint and the figure of the right viewpoint in the case where giving some parallax
The difference of two squares of two points.
(2) cost polymerization be exactly to limit in window Matching power flow a little sum.
(3) parallax optimization is then the parallax value that polymerization Least-cost is picked out to each point.
In addition, classical local algorithm also has absolute value summation algorithm (sum-of-absolute-
differences)：Matching power flow is the absolute value for calculating corresponding points；Adaptive weighting algorithm (adaptive-support-
window)：Color, the distance different weights of imparting i.e. in cost polymerization process according to the point in window from central point, then will
Its cost is added；Normalized crosscorrelation algorithm (Normalized Cross Correlation)：Calculate the cross-correlation between image
Value.Census is converted：The relative value of pixel grey scale size in window is changed into bit number, then compares the ratio of left and right corresponding points
The Hamming distance of special number.
Local algorithm has high speed and low energy consumption, but the matching precision that local algorithm is obtained is limited, it is more difficult to meet multiple
The demand of miscellaneous scene.Compared with local algorithm, Global Algorithm has more preferable matching precision.Global Algorithm is made for smooth item
It is assumed that and stereo matching problem is become into an energy-optimised problem.Most of Global Algorithm has skipped cost polymerization (step
2) energy function, it is considered to Matching power flow and smooth item, is proposed for global point, energy function is then minimized, depending on
Difference.And the optimized algorithm minimized to energy function mainly has figure to cut (graph cuts), confidence spread (Belief
Propagation) etc..But compared to local algorithm, the amount of calculation of Global Algorithm is bigger, and energy consumption is higher.
In actual applications, for especially for mobile device, the speed and precision of Stereo Matching Algorithm how to be balanced
It is exactly one of research emphasis, that is, the higher Real Time Matching Algorithm of the precision how to be realized in limited resource.Numerous
Among algorithm, half global registration (Semi-Global Match) be exactly showed after matching precision and operation time is balanced compared with
A kind of good algorithm, is widely used in a variety of applications.Half global registration algorithm proposes a global energy function, with overall situation calculation
Method optimizes difference to global point, and the energy function of each point is divided into the path in 16 directions of 8or by it, it is only necessary to right
Per paths evaluation, then the value addition in all paths is exactly the energy of the point, and can be adopted for the evaluation of single-pathway
Solved with Dynamic Programming.With the development of hardware, increasing hardware platform (GPU, FPGA etc.) is utilized to realize solid
Matching, particularly half global registration, to reach the purpose of real time execution.However, current algorithm is mostly just for static list
Two field picture is handled, it is not intended that the Stereo matching under video sequence.Do not have related work sutdy at present, and realize essence
It is still expensive that true real-time volume matches the cost for mobile platform.
The content of the invention
For above-mentioned the problems of the prior art, the invention discloses a kind of depth of utilization video sequence interframe correlation
Image computational methods, measure during the calculating for reducing Stereo matching, one or more above mentioned problems or shortcoming are solved in whole or in part.
The present invention is achieved through the following technical solutions, and the present invention comprises the following steps：
Step 1, left and right view camera are shot to Same Scene, the picture of shooting are corrected, after being corrected
Left and right different points of view identical size video sequence frame.
Step 2, for the first frame left and right picture point p, calculate Matching power flow in maximum disparity hunting zone, according to
The disparity search scope and cost value of p points carry out parallax optimization, according to what is obtained in the disparity search scope of p points and parallax optimization
The energy value of p points carries out parallax refinement；Obtain initial parallax value d1；The gradient of disparity of first frame is g1.
Step 3, the calculating for carrying out next frame, for p points：When p points are not abnormity point or marginal point, with previous frame p points
Disparity search scope is set on the basis of parallax value, Matching power flow is calculated；Otherwise, in matching generation, is calculated with maximum disparity hunting zone
Valency；Parallax optimization is carried out according to the disparity search scope of p points and cost value, optimized according to the disparity search scope of p points and parallax
In the obtained energy value of p points carry out parallax refinement；Obtain the parallax value and gradient of disparity of present frame；
Step 4, video sequence depth image is obtained, realize real-time Stereo matching.
Further, the energy value of each point in calculating process is retained, and it is follow-up to record the hunting zone of the point
Step parallax optimizes and parallax refinement is prepared.
Step 2 further comprises following technical characteristic：It is (0, D) to make disparity search scope, and wherein D is that disparity search is maximum
Value；For the left and right picture of the first frame, half global registration algorithm is taken to calculate initial parallax value, and retain each in calculating process
The energy value Energy [0] of individual point~Energy [D-1]；Wherein using census transformation calculations Matching power flow cost [0]~
Cost [D-1], accumulative optimization is carried out for energy function in 8 directions, is regarded for optimizing obtained original disparity map
Difference refinement.
Further, parallax refinement (post processing) process includes：Sub-pix refinement, singular values standard form rationalizes parallax etc.
Means.
Step 3 further comprises following technical characteristic：The calculating of next frame i+1 frame is carried out, when the cost value for calculating point p
When, find the parallax value di of previous frame the i-th frame p points, the point p of i+1 frame disparity search scope on the basis of di, from (0,
D) taper to (di-range, di+range), range is the hunting zone of setting.
Given threshold t1, t2, ask for the gradient map Gi of previous frame the i-th frame parallax, and two conditions are judged respectively：
(1) in cost calculating process, the cost value minimum value min (cost) calculated in (di-range, di+range)>
t1；
(2) gradient gis of the point p in Gi>t2；
The point for meeting condition (1) is abnormity point, and the point for meeting condition (2) is marginal point, as long as meeting one of them, is then expanded
Big current hunting zone is (0, D), returns and continues to calculate uncalculated cost value, and is recorded after the hunting zone of the point is
The continuous optimization of step parallax and parallax refinement are prepared.
Further, when carrying out the calculating of i+1 frame, for the point within the scope of disparity search, calculate to correspond to and be somebody's turn to do
Energy value in point range；For the point outside disparity search scope, the energy value outside the i-th frame point scope is directly inherited.
Further, in parallax thinning process, the thin of parallax only is carried out to the disparity search scope determined in step 3
Change, determine optimal parallax；The parallax value for making energy function minimum is found in the range of point p disparity search, the parallax is point p
Parallax.
The invention has the advantages that：1st, in real time application in, by the parallax obtained by previous frame image retain to work as
Previous frame.By retrieving the information of previous frame parallax, the hunting zone of present frame parallax is reduced, Matching power flow meter is in particular in
Calculate, in terms of parallax optimization and refinement.The algorithm improves whole algorithm and existed under conditions of original algorithmic match precision is ensured
Efficiency in practical application scene.2nd, to error increase marginal point, abnormity point, corresponding expansion disparity search scope are reduced
May caused error.3rd, retain energy value of each point in calculating process, each point in computational methods of the invention to regard
Poor hunting zone is that the optimization of subsequent step parallax and parallax refinement are prepared, and is reduced largely using video sequence interframe correlation
Amount of calculation.Therefore the calculating time of Stereo matching can be greatly decreased in the whole process of the present invention, while having ensured the essence of algorithm
Degree, has very big application prospect on real-time platform.
Brief description of the drawings
The structure that Fig. 1 matches for the depth image based on single-frame images.
Fig. 2 is the structure of the depth image computational methods of the embodiment of the present invention.
Fig. 3 is the algorithm flow of the start frame of the embodiment of the present invention.
Fig. 4 is the algorithm flow of the subsequent frame of the embodiment of the present invention.
Embodiment
For the ease of the understanding of those skilled in the art, the present invention is made further with reference to embodiment and accompanying drawing
It is bright.
When realizing Stereo matching in real time, acquisition is all a series of continuous picture frames, and is had perhaps between frame and frame
Many correlations, these correlations can be used to reduce the amount of calculation of whole algorithm.As shown in Fig. 2 the processing of the present embodiment
Step includes：
1st, the camera of left and right viewpoint is shot to Same Scene, is corrected for the picture of shooting, after being corrected
Left and right different points of view identical size picture.
2nd, it is (0, D) that might as well make disparity search scope, and wherein D is disparity search maximum.For the left and right figure of the first frame
Piece, takes unmodified half global registration algorithm to calculate initial parallax value, and retain the energy value of each point in calculating process
Energy [0]~Energy [D-1].Census transformation calculations Matching power flow cost [0]~cost [D-1] is wherein used, for
Energy function carries out accumulative optimization in 8 directions, and energy value Energy [0]~Energy is obtained using energy function optimization
[D-1], the original disparity map obtained for optimizing carries out parallax refinement.Parallax refinement (post processing) process mainly includes：Sub- picture
Plain refinement, singular values standard form rationalizes the means such as parallax.For modified hydrothermal process, parallax thinning process is constant.Details please see Figure
3。
3rd, the calculating of next frame is carried out, when calculating point p cost value, the parallax value d1 of previous frame p points is found, then
The point p of present frame disparity search scope tapers to (d1-range, d1+range), range then with d1 benchmark from (0, D)
For the disparity search scope of setting, 2~3 are typically taken.Dimension of picture is bigger, and range value is bigger.Matching power flow is calculated
(referred to herein as census conversion) calculates the cost value cost [d1-range] obtained in the range of (d1-range, d1+range)
~cost [d1+range].
4th, reducing matching range reduces the time of matching, but for the parallax value of edge, error can increase.Mainly exist
In：
(1), algorithm has error in itself, and the parallax of previous frame has error in itself, and the disparity search scope of present frame is base
Searched on the basis of the parallax value of previous frame, then more than the meeting of part parallax scan for, cause on the basis of 1 frame error point
As a result it is not inconsistent with truth.And the calculating of present frame can also produce new error.
(2), for the edge of object, the parallax value difference on edge both sides is larger.So for previous frame position in p points
For marginal point, it is moved in present frame marginal point at p+n, and the point p of present frame still one frame border point point p of above parallaxes
On the basis of search for parallax value, will result in error.
For error, given threshold t1, t2 ask for the gradient map G of previous frame parallax, two conditions are judged respectively
(1) in cost calculating process, the cost value minimum value min (cost) calculated in (d1-range, d1+range)>
t1。
(2) gradient g1s of the point p in G>t2.
Meet condition (1) thinks that the point is abnormity point, and meet condition (2) thinks that the point is marginal point, as long as meeting
One of them, then expand current hunting zone for (0, D), returns and continues to calculate uncalculated cost value.
If p points meet (1) or (2), the cost value not calculated is calculated according to the disparity search scope (0, D) expanded
[d1-range]s ＆cost [d1+range]~cost [D-1] thus obtains p points in disparity search scope to cost [0]~cost
The cost value cost [0] of (0, D)~cost [D-1], and it is the optimization of subsequent step parallax to update the disparity search scope of the point
Prepared with parallax refinement.If being unsatisfactory for decision condition (1) and (2), it is qualified to be considered as, and is directly entered next step, parallax
Optimization.
5th, it is exactly parallax optimization after cost is calculated.Hunting zone such as fruit dot p is (d1-range, d1+range), then
Calculate and update energy value Energy [d1-range-1]~Energy [d1+range-1], such as fruit dot p hunting zone for (0,
D), calculate and update energy value Energy [0]~Energy [D-1].Value not in the range of keeps constant.Then searched in point p
The parallax value for making energy function minimum is found in the range of rope, the parallax is point p parallax.Details see Fig. 4.
6th, in parallax refinement last handling process, the refinement of parallax is only carried out to the hunting zone that abovementioned steps are determined, really
Fixed optimal parallax.
In order to illustrate the improvement of whole service time.Assuming that having N number of point, disparity search scope is (0, D), then in cost
Accumulative, parallax optimizes, the former algorithm of parallax refinement aspect needs 3DN related operation altogether, and the algorithm after optimizing, and does not consider to calculate
If the gradient calculation for measuring very little, it is assumed that the ratio for needing to retain hunting zone D is n, and the scope after diminution is 2range+1, no
Consider the judgement in calculating process, then need 3* (2range+1) * (1-n) N+3*n*N*D related operation of *.For 640*
480, D=64, range=2, n=0.1 situation, then the amount of calculation of the latter account for the former 44%, it is contemplated that gradient calculation, meter
Calculation amount is 50% or so.
So the calculating time of Stereo matching can be greatly decreased in whole process, while the precision of algorithm has been ensured, real-time
There is very big application prospect on platform.
The technological thought of embodiment above only to illustrate the invention, it is impossible to which protection scope of the present invention is limited with this, it is all
It is any change done on the basis of technical scheme according to technological thought proposed by the present invention, each falls within present invention protection model
Within enclosing.The technology that the present invention is not directed to can be realized by existing technology.
Claims (7)
1. a kind of depth image computational methods of utilization video sequence interframe correlation, it is characterised in that comprise the following steps：
Step 1, left and right view camera are shot to Same Scene, and the picture of shooting is corrected, the left side after being corrected
The identical size video sequence frame of right different points of view；
Step 2, for the first frame left and right picture point p, in maximum disparity hunting zone, Matching power flow is calculated, according to p points
Disparity search scope and cost value carry out parallax optimization, according to the p points obtained in the disparity search scope of p points and parallax optimization
Energy value carries out parallax refinement；Obtain initial parallax value d1；The gradient of disparity of first frame is g1；
Step 3, the calculating for carrying out next frame, for p points：When p points are not abnormity point or marginal point, with the parallax of previous frame p points
Disparity search scope is set on the basis of value, Matching power flow is calculated；Otherwise, Matching power flow is calculated with maximum disparity hunting zone；Root
Parallax optimization is carried out according to the disparity search scope and cost value of p points, is obtained according in the disparity search scope of p points and parallax optimization
P points energy value carry out parallax refinement, obtain the parallax value and gradient of disparity of present frame；
Step 4, obtain video sequence depth image.
2. the depth image computational methods of utilization video sequence interframe correlation according to claim 1, it is characterised in that：
Retain the energy value of each point in calculating process；The disparity search scope of each point is recorded to optimize and regard for subsequent step parallax
Difference refinement is prepared.
3. the depth image computational methods of utilization video sequence interframe correlation according to claim 1, it is characterised in that
Step 2 further comprises following technical characteristic：
It is (0, D) to make disparity search scope, and wherein D is disparity search maximum；For the left and right picture of the first frame, use
Census transformation calculations Matching power flow cost [0]~cost [D-1], enter in 8 directions in parallax optimization for energy function
The accumulative optimization of row, obtains energy value Energy [0]~Energy [D-1], and the original disparity map obtained for optimizing carries out parallax
Refinement.
4. the depth image computational methods of utilization video sequence interframe correlation according to claim 1, it is characterised in that
The calculating Matching power flow of step 3 further comprises following technical characteristic：The calculating of next frame i+1 frame is carried out, when calculating point p's
During cost value, find the parallax value di of previous frame the i-th frame p points, the point p of i+1 frame disparity search scope on the basis of di,
Tapered to (di-range, di+range) from (0, D), range is the hunting zone of setting；
Given threshold t1, t2, ask for the gradient map Gi of previous frame the i-th frame parallax, and two conditions are judged respectively：
(1) in Matching power flow calculating process, the cost value minimum value min (cost) in (di-range, di+range)>t1；
(2) gradient gis of the point p in Gi>t2；
The point for meeting condition (1) is abnormity point, and the point for meeting condition (2) is marginal point, as long as meeting one of them, then expands and work as
Preceding hunting zone is (0, D), returns and continues to calculate uncalculated cost value.
5. the depth image computational methods of utilization video sequence interframe correlation according to claim 4, it is characterised in that：
When carrying out the calculating of i+1 frame, for the point within the scope of disparity search, the energy value corresponded in the point range is calculated；
For the point outside disparity search scope, the energy value outside the i-th frame point scope is directly inherited.
6. the depth image computational methods of utilization video sequence interframe correlation according to claim 4, it is characterised in that：
In parallax thinning process, the refinement of parallax is only carried out to the disparity search scope determined in step 3, optimal parallax is determined；
The parallax value for making energy function minimum is found in the range of point p disparity search, the parallax is point p parallax.
7. according to the depth image computational methods of any described utilization video sequence interframe correlation in claim 1 to 6, its
It is characterised by：Parallax thinning process includes sub-pix refinement, and singular values standard form rationalizes parallax.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710172103.XA CN106952304B (en) | 2017-03-22 | 2017-03-22 | A kind of depth image calculation method using video sequence interframe correlation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710172103.XA CN106952304B (en) | 2017-03-22 | 2017-03-22 | A kind of depth image calculation method using video sequence interframe correlation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106952304A true CN106952304A (en) | 2017-07-14 |
| CN106952304B CN106952304B (en) | 2019-09-20 |
Family
ID=59472179
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710172103.XA Active CN106952304B (en) | 2017-03-22 | 2017-03-22 | A kind of depth image calculation method using video sequence interframe correlation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106952304B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108876835A (en) * | 2018-03-28 | 2018-11-23 | 北京旷视科技有限公司 | Depth information detection method, device and system and storage medium |
| CN110033483A (en) * | 2019-04-03 | 2019-07-19 | 北京清微智能科技有限公司 | Based on DCNN depth drawing generating method and system |
| CN110148168A (en) * | 2019-05-23 | 2019-08-20 | 南京大学 | A kind of three mesh camera depth image processing methods based on the biradical line of size |
| CN110191330A (en) * | 2019-06-13 | 2019-08-30 | 内蒙古大学 | Depth map FPGA implementation method and system based on binocular vision green crop video flowing |
| CN110246169A (en) * | 2019-05-30 | 2019-09-17 | 华中科技大学 | A kind of window adaptive three-dimensional matching process and system based on gradient |
| CN111553296A (en) * | 2020-04-30 | 2020-08-18 | 中山大学 | Two-value neural network stereo vision matching method based on FPGA |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102026013A (en) * | 2010-12-18 | 2011-04-20 | 浙江大学 | Stereo video matching method based on affine transformation |
| CN103136750A (en) * | 2013-01-30 | 2013-06-05 | 广西工学院 | Stereo matching optimization method of binocular visual system |
| CN103810690A (en) * | 2012-11-07 | 2014-05-21 | 富士通株式会社 | Stereo matching method and device thereof |
| CN104065954A (en) * | 2014-07-03 | 2014-09-24 | 中国传媒大学 | Method for quickly detecting parallax scope of high-definition stereoscopic video |
| US20140321559A1 (en) * | 2013-04-24 | 2014-10-30 | Sony Corporation | Local detection model (ldm) for recursive motion estimation |
| CN106228605A (en) * | 2016-07-29 | 2016-12-14 | 东南大学 | A kind of Stereo matching three-dimensional rebuilding method based on dynamic programming |
-
2017
- 2017-03-22 CN CN201710172103.XA patent/CN106952304B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102026013A (en) * | 2010-12-18 | 2011-04-20 | 浙江大学 | Stereo video matching method based on affine transformation |
| CN103810690A (en) * | 2012-11-07 | 2014-05-21 | 富士通株式会社 | Stereo matching method and device thereof |
| CN103136750A (en) * | 2013-01-30 | 2013-06-05 | 广西工学院 | Stereo matching optimization method of binocular visual system |
| US20140321559A1 (en) * | 2013-04-24 | 2014-10-30 | Sony Corporation | Local detection model (ldm) for recursive motion estimation |
| CN104065954A (en) * | 2014-07-03 | 2014-09-24 | 中国传媒大学 | Method for quickly detecting parallax scope of high-definition stereoscopic video |
| CN106228605A (en) * | 2016-07-29 | 2016-12-14 | 东南大学 | A kind of Stereo matching three-dimensional rebuilding method based on dynamic programming |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108876835A (en) * | 2018-03-28 | 2018-11-23 | 北京旷视科技有限公司 | Depth information detection method, device and system and storage medium |
| CN110033483A (en) * | 2019-04-03 | 2019-07-19 | 北京清微智能科技有限公司 | Based on DCNN depth drawing generating method and system |
| CN110148168A (en) * | 2019-05-23 | 2019-08-20 | 南京大学 | A kind of three mesh camera depth image processing methods based on the biradical line of size |
| CN110246169A (en) * | 2019-05-30 | 2019-09-17 | 华中科技大学 | A kind of window adaptive three-dimensional matching process and system based on gradient |
| CN110191330A (en) * | 2019-06-13 | 2019-08-30 | 内蒙古大学 | Depth map FPGA implementation method and system based on binocular vision green crop video flowing |
| CN111553296A (en) * | 2020-04-30 | 2020-08-18 | 中山大学 | Two-value neural network stereo vision matching method based on FPGA |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106952304B (en) | 2019-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106952304B (en) | A kind of depth image calculation method using video sequence interframe correlation | |
| CN109271933B (en) | Method for estimating three-dimensional human body posture based on video stream | |
| CN110717527B (en) | Method for determining target detection model by combining cavity space pyramid structure | |
| US8953874B2 (en) | Conversion of monoscopic visual content using image-depth database | |
| CN109064514B (en) | Projection point coordinate regression-based six-degree-of-freedom pose estimation method | |
| CN109658445A (en) | Network training method, increment build drawing method, localization method, device and equipment | |
| CN111260661B (en) | Visual semantic SLAM system and method based on neural network technology | |
| US11367195B2 (en) | Image segmentation method, image segmentation apparatus, image segmentation device | |
| CN106651897B (en) | Parallax correction method based on super-pixel segmentation | |
| CN110688905A (en) | Three-dimensional object detection and tracking method based on key frame | |
| Dai et al. | Adaptive disparity candidates prediction network for efficient real-time stereo matching | |
| US20160117573A1 (en) | Method and apparatus for extracting feature correspondences from multiple images | |
| CN108961385B (en) | SLAM composition method and device | |
| CN111402412A (en) | Data acquisition method and device, equipment and storage medium | |
| CN104079800A (en) | Shaking preventing method for video image in video surveillance | |
| CN111553296B (en) | Two-value neural network stereo vision matching method based on FPGA | |
| US7602966B2 (en) | Image processing method, image processing apparatus, program and recording medium | |
| CN114419102B (en) | Multi-target tracking detection method based on frame difference time sequence motion information | |
| CN113034681B (en) | Three-dimensional reconstruction method and device for spatial plane relation constraint | |
| Zhu et al. | PairCon-SLAM: Distributed, online, and real-time RGBD-SLAM in large scenarios | |
| CN117726747A (en) | Three-dimensional reconstruction method, device, storage medium and equipment for complementing weak texture scene | |
| CN112233149A (en) | Scene flow determination method and device, storage medium and electronic device | |
| CN116704112A (en) | 3D scanning system for object reconstruction | |
| KR101178015B1 (en) | Generating method for disparity map | |
| Jung et al. | Depth map refinement using super-pixel segmentation in multi-view systems |
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 |