CN106550229A - A kind of parallel panorama camera array multi-view image bearing calibration - Google Patents
A kind of parallel panorama camera array multi-view image bearing calibration Download PDFInfo
- Publication number
- CN106550229A CN106550229A CN201610904649.5A CN201610904649A CN106550229A CN 106550229 A CN106550229 A CN 106550229A CN 201610904649 A CN201610904649 A CN 201610904649A CN 106550229 A CN106550229 A CN 106550229A
- Authority
- CN
- China
- Prior art keywords
- point
- view image
- camera array
- panorama camera
- parallel
- 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.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
- H04N13/10—Processing, recording or transmission of stereoscopic or multi-view image signals
- H04N13/106—Processing image signals
- H04N13/128—Adjusting depth or disparity
Abstract
The invention discloses a kind of parallel panorama camera array multi-view image bearing calibration, comprises the following steps:S1:Extract and match each visual point image characteristic point:Enough SIFT feature points to be matched are detected from the image of multiple viewpoints, position and the yardstick of characteristic point is determined;S2:Key point is accurately positioned:To the fitting of feature candidate point ambient data that determines trying to achieve accurate position, yardstick and curvature ratio;S3:Multi-view image adjacent two-by-two is corrected according to match point information after selected becoming more meticulous;S4:Projection multi-view image is to public correcting plane;S5:The generation of description vector;S6:Characteristic matching:It is determined that the character pair point relation of adjacent image centering two-by-two;S7:The matching characteristic point extracted by each stereo pair using improved RANSAC become more meticulous algorithm remove noise spot.The present invention is reasonable in design, and characteristic matching ability is strong, is conducive to improving the precision of parametric solution and the panorama camera array multi-view image corrects efficiency.
Description
Technical field
The present invention relates to multi-view image bearing calibration technical field, more particularly to regard a kind of parallel panorama camera array more
Dot image bearing calibration.
Background technology
Polyphaser Real-time Collection three-dimensional display system is an important application of three-dimensional television, and multiple cameras are adopted by real time
The video of collection is sent to server, and in the server multi-view image is processed, and solid is displayed in after synthetic stereo image
On display.For run-in index camera array, as inevitable operating error and equipment precision are limited, only with craft
The method of adjustment camera position can not simulate putting for ideal parallelism formula camera array completely.Using this manual placement camera
The multi-view image that mode shoots, character pair point are unjustified in vertical direction, and parallax is uneven in the horizontal direction, so as to
In causing the stereo-picture for synthesizing, there is obvious scintillation in object, has had a strong impact on the viewing effect of real-time three-dimensional three-dimensional video-frequency
Really.RANSAC algorithms multi-view image correct with being to calculate number according to one group of sample data set comprising abnormal data
According to mathematical model parameter, so as to obtain the algorithm of effective sample data, traditional RANSAC algorithms are in terms of noise spot is removed
There is good performance, but it is less efficient.
The content of the invention
The invention aims to shortcoming present in prior art is solved, and a kind of parallel panorama camera battle array for proposing
Row multi-view image bearing calibration.
To achieve these goals, present invention employs following technical scheme:
A kind of parallel panorama camera array multi-view image bearing calibration, comprises the following steps:
S1:Extract and match each visual point image characteristic point:Enough SIFT feature points to be matched are detected from the image of multiple viewpoints,
Determine position and the yardstick of characteristic point, stable feature, metric space letter are searched in all of yardstick with metric space function
Number is expressed as:;
S2:Key point is accurately positioned:To determine feature candidate point ambient data fitting come try to achieve accurate position, yardstick and
Curvature ratio, abandons that neighborhood contrast is relatively low or candidate point on edge, and principal curvatures is by calculatingHessian matrixesTry to achieve:;
S3:Multi-view image adjacent two-by-two is corrected according to match point information after selected becoming more meticulous;
S4:Projection multi-view image is to public correcting plane:According to each stereo image correction matrix for calculating, it is determined that will be many
Visual point image is projected to the method for public correcting plane;
S5:The generation of description vector:First by the coordinate and gradient direction of each sampled point in neighborhood before description vector is generated
A rotation is done all along the principal direction of key point, different weights, Gauss are assigned to each sampled point with Gauss weighting function
In key point position, variance is the half of crucial vertex neighborhood window width at the center of function;
S6:Characteristic matching:Enough SIFT features to be matched are detected from the image of multiple viewpoints, it is determined that adjacent image pair two-by-two
In character pair point relation;
S7:The matching characteristic point extracted by each stereo pair using improved RANSAC become more meticulous algorithm remove noise spot:Often
3 matchings of secondary sampling are to calculating this transformation matrix model, then bringing each point into the modeling statistics and meeting in the model
Count out, using the prevalence splicing of video-splicing model, after crucial frame sampling, based on translational motion, then with simple
The method of average and standard deviation calculates offset data in x filtering noise data, average x_mean in y-axis direction, y_mean,
Offset data is calculated in x, y, axial standard deviation x_std, y_std, set S are used as new RANSAC data, compound stereoscopic
Image, wherein:。
Preferably, in the S1, symbolRepresent convolution algorithm, gaussian kernelFor:。
Preferably, in the S1, imageMetric space functionGaussian kernel comprising yardstick variableWith imageConvolution, i.e.,:。
Preferably, in the S1, the normalized LOG operators of metric space function replace DOG, the normalized LOG
Operator is, will obtain:。
Preferably, in the S1, willEquation two ends with divided by denominator,
To obtain:。
Preferably, in the S1, two neighboring yardstickWithOnFinite difference is come approximate,
To obtain:。
Preferably, in the S2, with Taylor expansion by metric space functionExpansion has:。
Preferably, in the S2,To derivation and make derivative be zero, side-play amount will be obtainedFor:。
Preferably, in the S2, by side-play amountBring intoIn be obtained:。
The invention has the beneficial effects as follows:Multi-view image is projected to public correcting plane by projective transformation, facilitates level
Freely adjusting for parallax, effectively increases multi-view image correction use range, and characteristic matching ability is strong, in picture quality
Also can normal work in the case of poor;Instead of using DOG, it is to avoid solution second derivative, substantially increase meter
Calculate speed;And algorithm is purified using improved RANSAC, the noise ratio of data is greatly reduced again, improves RANSAC results
The purity of reliability and data, is conducive to the precision and the panorama camera array multi-view image of parametric solution to correct efficiency.This
Invention is reasonable in design, and characteristic matching ability is strong, increased multi-view image correction use range, is conducive to improving parametric solution
Precision and the panorama camera array multi-view image correct efficiency.
Specific embodiment
The present invention is further explained with reference to specific embodiment.
Embodiment
The present embodiment proposes a kind of parallel panorama camera array multi-view image bearing calibration, comprises the following steps:
S1:Extract and match each visual point image characteristic point:Enough SIFT feature points to be matched are detected from the image of multiple viewpoints,
Determine position and the yardstick of characteristic point, stable feature, metric space letter are searched in all of yardstick with metric space function
Number is expressed as:;
S2:Key point is accurately positioned:To determine feature candidate point ambient data fitting come try to achieve accurate position, yardstick and
Curvature ratio, abandons that neighborhood contrast is relatively low or candidate point on edge, and principal curvatures is by calculatingHessian matrixesTry to achieve:;
S3:Multi-view image adjacent two-by-two is corrected according to match point information after selected becoming more meticulous;
S4:Projection multi-view image is to public correcting plane:According to each stereo image correction matrix for calculating, it is determined that will be many
Visual point image is projected to the method for public correcting plane;
S5:The generation of description vector:First by the coordinate and gradient direction of each sampled point in neighborhood before description vector is generated
A rotation is done all along the principal direction of key point, different weights, Gauss are assigned to each sampled point with Gauss weighting function
In key point position, variance is the half of crucial vertex neighborhood window width at the center of function;
S6:Characteristic matching:Enough SIFT features to be matched are detected from the image of multiple viewpoints, it is determined that adjacent image pair two-by-two
In character pair point relation;
S7:The matching characteristic point extracted by each stereo pair using improved RANSAC become more meticulous algorithm remove noise spot:Often
3 matchings of secondary sampling are to calculating this transformation matrix model, then bringing each point into the modeling statistics and meeting in the model
Count out, using the prevalence splicing of video-splicing model, after crucial frame sampling, based on translational motion, then with simple
The method of average and standard deviation calculates offset data in x filtering noise data, average x_mean in y-axis direction, y_mean,
Offset data is calculated in x, y, axial standard deviation x_std, y_std, set S are used as new RANSAC data, compound stereoscopic
Image, wherein:。
Wherein symbolRepresent convolution algorithm, gaussian kernelFor:。
ImageMetric space functionGaussian kernel comprising yardstick variableWith imageVolume
Product, i.e.,:.Metric space function replaces DOG, the normalization with normalized LOG operators
LOG operators be, will obtain:.WillDeng
Formula two ends will be obtained with divided by denominator:.Two neighboring yardstickWith
OnFinite difference is come approximate, will obtain:。
With Taylor expansion by metric space functionExpansion has:。To derivation and make derivative be zero, side-play amount will be obtainedFor:.By side-play amountBring into
In be obtained:.Multi-view image is projected to public correcting plane by projective transformation, facilitates water
Look squarely it is poor freely adjust, effectively increase multi-view image correction use range, and characteristic matching ability be strong, in image matter
Also can normal work in the case that amount is poor;Instead of using DOG, it is to avoid solution second derivative, substantially increase
Calculating speed;And algorithm is purified using improved RANSAC, the noise ratio of data is greatly reduced again, improves RANSAC results
Reliability and data purity, be conducive to improving precision and the panorama camera array multi-view image correction effect of parametric solution
Rate.
The above, the only present invention preferably specific embodiment, but protection scope of the present invention is not limited thereto,
Any those familiar with the art the invention discloses technical scope in, technology according to the present invention scheme and its
Inventive concept equivalent or change in addition, should all be included within the scope of the present invention.
Claims (9)
1. a kind of parallel panorama camera array multi-view image bearing calibration, it is characterised in that comprise the following steps:
S1:Extract and match each visual point image characteristic point:Enough SIFT feature points to be matched are detected from the image of multiple viewpoints,
Determine position and the yardstick of characteristic point, stable feature, metric space letter are searched in all of yardstick with metric space function
Number is expressed as:;
S2:Key point is accurately positioned:To determine feature candidate point ambient data fitting come try to achieve accurate position, yardstick and
Curvature ratio, abandons that neighborhood contrast is relatively low or candidate point on edge, and principal curvatures is by calculatingHessian matrixesTry to achieve:;
S3:Multi-view image adjacent two-by-two is corrected according to match point information after selected becoming more meticulous;
S4:Projection multi-view image is to public correcting plane:According to each stereo image correction matrix for calculating, it is determined that will be many
Visual point image is projected to the method for public correcting plane;
S5:The generation of description vector:First by the coordinate and gradient direction of each sampled point in neighborhood before description vector is generated
A rotation is done all along the principal direction of key point, different weights, Gauss are assigned to each sampled point with Gauss weighting function
In key point position, variance is the half of crucial vertex neighborhood window width at the center of function;
S6:Characteristic matching:Enough SIFT features to be matched are detected from the image of multiple viewpoints, it is determined that adjacent image pair two-by-two
In character pair point relation;
S7:The matching characteristic point extracted by each stereo pair using improved RANSAC become more meticulous algorithm remove noise spot:Often
3 matchings of secondary sampling are to calculating this transformation matrix model, then bringing each point into the modeling statistics and meeting in the model
Count out, using the prevalence splicing of video-splicing model, after crucial frame sampling, based on translational motion, then with simple
The method of average and standard deviation calculates offset data in x filtering noise data, average x_mean in y-axis direction, y_mean,
Offset data is calculated in x, y, axial standard deviation x_std, y_std, set S are used as new RANSAC data, compound stereoscopic
Image, wherein:。
2. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1, it is characterised in that institute
State in S1, symbolRepresent convolution algorithm, gaussian kernelFor:。
3. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1 and 2, its feature exist
In, in the S1, imageMetric space functionGaussian kernel comprising yardstick variableWith figure
PictureConvolution, i.e.,:。
4. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1, it is characterised in that institute
State in S1, the normalized LOG operators of metric space function replace the DOG, the normalized LOG operators to be, will
Arrive:。
5. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1 or 4, its feature exist
In, in the S1, willEquation two ends will be obtained with divided by denominator:。
6. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1 or 5, its feature exist
In, in the S1, two neighboring yardstickWithOnFinite difference is come approximate, will obtain:。
7. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1, it is characterised in that institute
State in S2, with Taylor expansion by metric space functionExpansion has:。
8. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1 or 7, its feature exist
In, in the S2,To derivation and make derivative be zero, side-play amount will be obtainedFor:。
9. a kind of parallel panorama camera array multi-view image bearing calibration according to claim 1 or 8, its feature exist
In in the S2, by side-play amountBring intoIn be obtained:。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610904649.5A CN106550229A (en) | 2016-10-18 | 2016-10-18 | A kind of parallel panorama camera array multi-view image bearing calibration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610904649.5A CN106550229A (en) | 2016-10-18 | 2016-10-18 | A kind of parallel panorama camera array multi-view image bearing calibration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106550229A true CN106550229A (en) | 2017-03-29 |
Family
ID=58369153
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610904649.5A Pending CN106550229A (en) | 2016-10-18 | 2016-10-18 | A kind of parallel panorama camera array multi-view image bearing calibration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106550229A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107622510A (en) * | 2017-08-25 | 2018-01-23 | 维沃移动通信有限公司 | A kind of information processing method and device |
CN109348125A (en) * | 2018-10-31 | 2019-02-15 | Oppo广东移动通信有限公司 | Video correction method, apparatus, electronic equipment and computer readable storage medium |
CN109462746A (en) * | 2018-10-26 | 2019-03-12 | 北京双髻鲨科技有限公司 | A kind of method and device of image debounce |
WO2019096016A1 (en) * | 2017-11-14 | 2019-05-23 | 深圳岚锋创视网络科技有限公司 | Method for achieving bullet time capturing effect and panoramic camera |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102065313A (en) * | 2010-11-16 | 2011-05-18 | 上海大学 | Uncalibrated multi-viewpoint image correction method for parallel camera array |
CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
CN103559711A (en) * | 2013-11-05 | 2014-02-05 | 余洪山 | Motion estimation method based on image features and three-dimensional information of three-dimensional visual system |
CN104134200A (en) * | 2014-06-27 | 2014-11-05 | 河海大学 | Mobile scene image splicing method based on improved weighted fusion |
CN104318548A (en) * | 2014-10-10 | 2015-01-28 | 西安电子科技大学 | Rapid image registration implementation method based on space sparsity and SIFT feature extraction |
-
2016
- 2016-10-18 CN CN201610904649.5A patent/CN106550229A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102065313A (en) * | 2010-11-16 | 2011-05-18 | 上海大学 | Uncalibrated multi-viewpoint image correction method for parallel camera array |
CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
CN103559711A (en) * | 2013-11-05 | 2014-02-05 | 余洪山 | Motion estimation method based on image features and three-dimensional information of three-dimensional visual system |
CN104134200A (en) * | 2014-06-27 | 2014-11-05 | 河海大学 | Mobile scene image splicing method based on improved weighted fusion |
CN104318548A (en) * | 2014-10-10 | 2015-01-28 | 西安电子科技大学 | Rapid image registration implementation method based on space sparsity and SIFT feature extraction |
Non-Patent Citations (1)
Title |
---|
王莎: "基于特征的X线图像拼接算法研究与实现", 《万方数据》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107622510A (en) * | 2017-08-25 | 2018-01-23 | 维沃移动通信有限公司 | A kind of information processing method and device |
WO2019096016A1 (en) * | 2017-11-14 | 2019-05-23 | 深圳岚锋创视网络科技有限公司 | Method for achieving bullet time capturing effect and panoramic camera |
US11233944B2 (en) | 2017-11-14 | 2022-01-25 | Arashi Vision Inc. | Method for achieving bullet time capturing effect and panoramic camera |
CN109462746A (en) * | 2018-10-26 | 2019-03-12 | 北京双髻鲨科技有限公司 | A kind of method and device of image debounce |
CN109348125A (en) * | 2018-10-31 | 2019-02-15 | Oppo广东移动通信有限公司 | Video correction method, apparatus, electronic equipment and computer readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108564617B (en) | Three-dimensional reconstruction method and device for multi-view camera, VR camera and panoramic camera | |
CN109272570B (en) | Space point three-dimensional coordinate solving method based on stereoscopic vision mathematical model | |
CN102065313B (en) | Uncalibrated multi-viewpoint image correction method for parallel camera array | |
US9094672B2 (en) | Stereo picture generating device, and stereo picture generating method | |
CN106550229A (en) | A kind of parallel panorama camera array multi-view image bearing calibration | |
CN103854291B (en) | Camera marking method in four-degree-of-freedom binocular vision system | |
CN110853151B (en) | Video-based three-dimensional point set recovery method | |
CN107798702B (en) | Real-time image superposition method and device for augmented reality | |
CN107665483B (en) | Calibration-free convenient monocular head fisheye image distortion correction method | |
CN102436660A (en) | Automatic correction method and device of 3D camera image | |
CN103337094A (en) | Method for realizing three-dimensional reconstruction of movement by using binocular camera | |
CN109146794A (en) | A kind of light field image rotation error bearing calibration | |
CN107358633A (en) | Join scaling method inside and outside a kind of polyphaser based on 3 points of demarcation things | |
CN109308719A (en) | A kind of binocular parallax estimation method based on Three dimensional convolution | |
CN106127701A (en) | Fisheye image distortion correction method and device | |
CN106204429A (en) | A kind of method for registering images based on SIFT feature | |
CN104240229B (en) | A kind of adaptive method for correcting polar line of infrared binocular camera | |
CN112258588A (en) | Calibration method and system of binocular camera and storage medium | |
CN107016707B (en) | A kind of integration imaging super large three-dimensional scenic shooting method for correcting image | |
CN110099268B (en) | Blind area perspective display method with natural color matching and natural display area fusion | |
CN109992809A (en) | A kind of construction method of buildings model, device and storage device | |
CN103997638A (en) | Matrix type camera array multi-view image correction method | |
TW202113551A (en) | Virtuality reality overlapping method and system | |
CN104036496B (en) | Self-calibration method for radial distortion of fish-eye lens camera | |
CN111047636B (en) | Obstacle avoidance system and obstacle avoidance method based on active infrared binocular vision |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170329 |