CN105184856A - Two-phase human skin three-dimensional reconstruction method based on density matching - Google Patents
Two-phase human skin three-dimensional reconstruction method based on density matching Download PDFInfo
- Publication number
- CN105184856A CN105184856A CN201510554937.8A CN201510554937A CN105184856A CN 105184856 A CN105184856 A CN 105184856A CN 201510554937 A CN201510554937 A CN 201510554937A CN 105184856 A CN105184856 A CN 105184856A
- Authority
- CN
- China
- Prior art keywords
- image
- matching
- point
- human body
- body skin
- 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
Abstract
The invention discloses a two-phase human skin three-dimensional reconstruction method based on density matching. The method comprises carrying out registering rectification processing on image sequences in advance to detect all characteristic angle points in the image sequences after correction as texture characteristic points, searching for corresponding characteristic points along a polar line in a search scope set in an image by respectively taking the texture characteristic points as seeds, calculating similarity between the corresponding characteristic points, finishing sparsity matching, performing image triangulation by taking the characteristic points after the matching as summits to divide the image into a plurality of triangular areas, calculating a maximum parallax value of a matched area, calculating energy similarity values of the characteristic points, carrying out characteristic point matching on the corresponding triangular areas between the image sequences, finishing the density matching, calculating an external parameter matrix of a camera, and outputting a three-dimensional point cloud model. According to the invention, three-dimensional reconstruction is carried out on human skin by use of a common fixed focus camera, the operation is simple, the cost is quite low, and by use of the method provided by the invention, the point cloud model can be rapidly and efficiently obtained.
Description
Technical field
The present invention relates to computer vision and digital image processing field, be more particularly to the human body skin three-dimensional rebuilding method based on multi views dense matching.
Background technology
Current area of computer aided visualization system has been widely used in social each field, along with the excellent properties performance of computing machine in processing graphics image, user proposes higher demand to the real virtual environment of drafting, namely visual human is more true to nature better, and an important problem is the modeling of human body and the visual of model here.The visual of human body skin makes the key link that visual human obtains three-dimensional model, can be used for the fields such as online skin diagnosis, development of games, video display special efficacy, augmented reality.
Spatial digitizer is the method for acquisition object dimensional data comparatively common in practical application, in the application of Human Modeling, because laser is to the potential impact of human body skin, and the general method adopted based on image modeling.Based on the mode of the stereoscopic vision direct modeling human vision process scenery of multi views, the three-dimensional data of scenery can be obtained under numerous conditions quickly and easily, its effect is that other visible sensation methods can not replace, to its research, be from the angle of vision physiological or engineer applied all tool be of great significance, " TwoStagesStereoDenseMatchingAlgorithmfor3DSkinMicro-Surf aceReconstruction " (Zhang Qian, " IEEE core database ", the online publishing date: on January 9th, 2010) describe the method for human body skin method based on multi views dense matching and three-dimensional reconstruction in a literary composition in detail, in the mode introduced, its full content is introduced herein.
Modeling based on image is a kind of data-driven model, under given skin condition, fixed-focus digital camera is utilized to take image sequence, between image, character pair point is for judging the volume coordinate of object corresponding point, recover three-dimensional cutaneous grid by camera parameter matrix, the detection of modeling process unique point and the coupling of point of interest are key issues to be solved.Non-rigid is human body skin particularly, affects color and brightness can produce larger change by light, when when there is space deformation and the uneven situation of local skin grain distribution, carrying out coupling become quite difficulty to dense characteristic point set to be matched.Under the steric prerequisite of consideration dermatoglyph characteristic sum coupling point set, be necessary to invent the dense matching problem that a kind of new model method solves skin or non-rigid object.But this field maturation method of being correlated with is less at present, and the general dense matching method based on multi views can not directly apply to the three-dimensional reconstruction of skin, and the effect after reconstruction is undesirable, and efficiency is lower.
Summary of the invention
The object of the invention is to overcome the deficiencies in the prior art, provide a kind of two-stage human body skin three-dimensional rebuilding method based on dense matching, its disposal route is with strong points, easy to operate, rapidly and efficiently, to promote that human body skin modeling and emotion play up the deep application in industries such as video display animation, game, medical diagnosis, safety certifications.
Based on the two-stage human body skin three-dimensional rebuilding method of dense matching, comprise the following steps successively:
(1) adopt fixed-focus digital camera capturing skin image sequence, carry out the registration correction process between image;
(2) Corner Detection is carried out to image sequence;
(3) setting search window w × w, for seed, searches the match point of corresponding correspondence with wherein piece image feature angle point in other n-1 sub-picture;
(4) calculate the similarity between match point, calculate the Corresponding matching point of all angle points successively, complete sparse coupling;
(5) with feature angle point for summit, partitioned image region becomes some adjacent triangle, and the maximum disparity of each delta-shaped region is three summit parallax (V
1, V
2, V
3) maximal value, disparity range is decided to be { 0, max (V
1, V
2, V
3);
(6) in feature trigonometric ratio region according to region increase mode, calculate the energy similarity measure between feature point pairs, the match is successful to be less than defined threshold, carry out successively coupling a little, complete dense matching, energy similarity measure calculates and carries out according to following formula:
Wherein, d is disparity range maximal value, and c is R, G, B format set of coloured image, is respectively that { 1,2,3}, l and r represent the pixel value of spatially left image and right image respectively, x and y is pixel x direction and the y direction coordinate of image;
(7) the outer parameter matrix of camera is calculated;
(8) calculate the volume coordinate of all unique points, export three-dimensional point cloud model.
Preferably, in described step (1), the registration antidote of image sequence is the antidote of non-scalable video.
Preferably, in described step (2), angular-point detection method is Harris corner detection approach.
Preferably, in described step (3), the resolution of search window is set to 15 × 15 or 20 × 20.
Preferably, the matching process in described step (4) adopts RANSAC algorithm RANSAC, and wherein maximum disparity is decided to be 60 ~ 90 pixels.
Preferably, described step (5) intermediate cam method is the triangulation of Delaunay point set.
Preferably, in described step (6), energy similarity measure threshold value is set as the numerical value between 10 ~ 20.
Accompanying drawing explanation
Fig. 1 is based on the two-stage human body skin three-dimensional rebuilding method process flow diagram of dense matching
The trigonometric ratio figure of feature angle point after the sparse coupling of Fig. 2 one-phase
The point cloud chart exported after Fig. 3 two-stage dense matching
Three-dimensional reconstruction design sketch after Fig. 4 texture mapping
Embodiment
Based on the skin pore identifying processing method of graphical analysis, comprise the following steps successively:
(1) fixed-focus digital camera capturing skin image sequence is adopted, carry out the registration correction process between image, determine the internal reference matrix 3*3 of every width image, calculate the collinear transformation matrix of every sub-picture, calculate the position coordinates of every sub-picture center pixel, internal reference matrix and collinear transformation matrix multiple, obtain new internal reference matrix, the geometric transformation of computed image, obtain the polar curve equation of image, every sub-picture matrix is converted along polar curve direction, obtains the image sequence after polar curve calibration;
(2) carry out Corner Detection to image sequence, the difference operator of definition horizontal direction and vertical direction, is respectively {-1,0,1 ,-1,0,1,-1,0,1}, {-1,-1 ,-1,0,0,0,1,1,1}, image makes convolution with the difference operator of both direction respectively, selects the window of 5 × 5 to make Gaussian smoothing, calculates the angle point response of each pixel, obtains the angle point of image;
(3) setting search window w × w, for seed, searches the match point of corresponding correspondence with wherein piece image feature angle point in other n-1 sub-picture;
(4) calculate the pixel value between match point and point to be matched square, as the similarity between match point, similarity, higher than the point that is that the match is successful of setting threshold value, calculates the Corresponding matching point of all angle points successively, completes sparse coupling;
(5) with feature angle point for summit, image-region is divided into some adjacent triangle, and the maximum disparity of each delta-shaped region is this triangular apex parallax (V
1, V
2, V
3) maximal value, disparity range is decided to be { 0, max (V
1, V
2, V
3);
(6) in feature trigonometric ratio region according to the mode that region increases, with feature angle point for seed, calculate the energy similarity measure between feature point pairs, the match is successful to be less than defined threshold, carry out successively coupling a little, complete dense matching, energy similarity measure calculate carry out according to following formula:
Wherein, d is disparity range maximal value, and c is R, G, B format set of coloured image, is respectively that { 1,2,3}, l and r represent the pixel value of spatially left image and right image respectively, x and y is pixel x direction and the y direction coordinate of image;
(7) utilize the sparse point of the coupling in step (4) right, calculate perspective projection matrix, decomposed the rotation matrix and translation matrix that obtain in outer parameter matrix by perspective projection matrix;
(8) outer parameter matrix is multiplied by the volume coordinate that match point coordinates matrix obtains image, calculates the volume coordinate obtaining all match points successively, exports three-dimensional point cloud model.
Preferably, in described step (1), the registration antidote of image sequence is the antidote of non-scalable video.
Preferably, in described step (2), angular-point detection method is Harris corner detection approach.
Preferably, in described step (3), the resolution of search window is set to 15 × 15 or 20 × 20.
Preferably, the matching process in described step (4) adopts RANSAC algorithm RANSAC, and wherein maximum disparity is decided to be 60 ~ 90 pixels.
Preferably, described step (5) intermediate cam method is the triangulation of Delaunay point set.
Preferably, in described step (6), energy similarity measure threshold value is set as the numerical value between 10 ~ 20.
This method is with strong points, does not need spatial digitizer, and cost is low, easy to operate, rapidly and efficiently, can in visual human's modeling Rapid Popularization.
Although for illustrative purposes; describe illustrative embodiments of the present invention; but it should be appreciated by those skilled in the art that; when not departing from scope of invention disclosed in claims and spirit; the change of various amendment, interpolation and replacement etc. can be carried out in form and details; and all these change the protection domain that all should belong to claims of the present invention; and application claims protection each department of product and method in each step, can combine with the form of combination in any.Therefore, be not intended to limit the scope of the invention to the description of embodiment disclosed in the present invention, but for describing the present invention.Correspondingly, scope of the present invention not by the restriction of above embodiment, but is limited by claim or its equivalent.
Claims (7)
1., based on a two-stage human body skin three-dimensional rebuilding method for dense matching, comprise the following steps successively:
(1) adopt fixed-focus digital camera capturing skin image sequence, carry out the registration correction process between image;
(2) Corner Detection process is carried out to image sequence;
(3) setting search window w × w, for seed, searches the match point of corresponding correspondence with wherein piece image feature angle point in other n-1 sub-picture;
(4) calculate the similarity between match point, calculate the Corresponding matching point of all angle points successively, complete sparse coupling;
(5) with feature angle point for summit, partitioned image region becomes some adjacent triangle, and the maximum disparity of each delta-shaped region is three summit parallax (V
1, V
2, V
3) maximal value, disparity range is decided to be { 0, max (V
1, V
2, V
3);
(6) in feature trigonometric ratio region according to region increase mode, calculate the energy similarity measure between feature point pairs, the match is successful to be less than defined threshold, carry out successively coupling a little, complete dense matching, energy similarity measure calculates and carries out according to following formula:
Wherein, d is disparity range maximal value, and c is R, G, B format set of coloured image, is respectively that { 1,2,3}, l and r represent the pixel value of spatially left image and right image respectively, x and y is pixel x direction and the y direction coordinate of image;
(6) the outer parameter matrix of camera is calculated;
(7) calculate the volume coordinate of all unique points, export three-dimensional point cloud model.
2., as claimed in claim 1 based on the two-stage human body skin three-dimensional rebuilding method of dense matching, it is characterized in that: the registration antidote of image sequence is the antidote of non-scalable video.
3., as claimed in claim 2 based on the two-stage human body skin three-dimensional rebuilding method of dense matching, it is characterized in that: angular-point detection method is Harris corner detection approach.
4., as claimed in claim 3 based on the two-stage human body skin three-dimensional rebuilding method of dense matching, it is characterized in that: the resolution of search window is set to 15 × 15 or 20 × 20.
5. as claimed in claim 4 based on the two-stage human body skin three-dimensional rebuilding method of dense matching, it is characterized in that: matching process adopts RANSAC algorithm RANSAC, and wherein maximum disparity is decided to be 60 ~ 90 pixels.
6., as claimed in claim 5 based on the two-stage human body skin three-dimensional rebuilding method of dense matching, it is characterized in that: Triangulation Algorithm is the triangulation of Delaunay point set.
7., as claimed in claim 1 based on the two-stage human body skin three-dimensional rebuilding method of dense matching, it is characterized in that: energy similarity measure threshold value is set as the numerical value between 10 ~ 20.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510554937.8A CN105184856A (en) | 2015-09-02 | 2015-09-02 | Two-phase human skin three-dimensional reconstruction method based on density matching |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510554937.8A CN105184856A (en) | 2015-09-02 | 2015-09-02 | Two-phase human skin three-dimensional reconstruction method based on density matching |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105184856A true CN105184856A (en) | 2015-12-23 |
Family
ID=54906906
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510554937.8A Pending CN105184856A (en) | 2015-09-02 | 2015-09-02 | Two-phase human skin three-dimensional reconstruction method based on density matching |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105184856A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109829502A (en) * | 2019-02-01 | 2019-05-31 | 辽宁工程技术大学 | It is a kind of towards repeating the picture of texture and non-rigid shape deformations to efficient dense matching method |
CN110021039A (en) * | 2018-11-15 | 2019-07-16 | 山东理工大学 | The multi-angle of view material object surface point cloud data initial registration method of sequence image constraint |
CN110245674A (en) * | 2018-11-23 | 2019-09-17 | 浙江大华技术股份有限公司 | Template matching method, device, equipment and computer storage medium |
CN110543871A (en) * | 2018-09-05 | 2019-12-06 | 天目爱视(北京)科技有限公司 | point cloud-based 3D comparison measurement method |
CN112734652A (en) * | 2020-12-22 | 2021-04-30 | 同济大学 | Near-infrared blood vessel image projection correction method based on binocular vision |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120313937A1 (en) * | 2010-01-18 | 2012-12-13 | Disney Enterprises, Inc. | Coupled reconstruction of hair and skin |
-
2015
- 2015-09-02 CN CN201510554937.8A patent/CN105184856A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120313937A1 (en) * | 2010-01-18 | 2012-12-13 | Disney Enterprises, Inc. | Coupled reconstruction of hair and skin |
Non-Patent Citations (1)
Title |
---|
QIAN ZHANG ET AL: "《Two Stages Stereo Dense Matching Algorithm for 3D Skin Micro-surface Reconstruction》", 《INTERNATIONAL CONFERENCE ON MULTIMEDIA MODELING》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543871A (en) * | 2018-09-05 | 2019-12-06 | 天目爱视(北京)科技有限公司 | point cloud-based 3D comparison measurement method |
CN110543871B (en) * | 2018-09-05 | 2022-01-04 | 天目爱视(北京)科技有限公司 | Point cloud-based 3D comparison measurement method |
CN110021039A (en) * | 2018-11-15 | 2019-07-16 | 山东理工大学 | The multi-angle of view material object surface point cloud data initial registration method of sequence image constraint |
CN110245674A (en) * | 2018-11-23 | 2019-09-17 | 浙江大华技术股份有限公司 | Template matching method, device, equipment and computer storage medium |
CN110245674B (en) * | 2018-11-23 | 2023-09-15 | 浙江大华技术股份有限公司 | Template matching method, device, equipment and computer storage medium |
CN109829502A (en) * | 2019-02-01 | 2019-05-31 | 辽宁工程技术大学 | It is a kind of towards repeating the picture of texture and non-rigid shape deformations to efficient dense matching method |
CN109829502B (en) * | 2019-02-01 | 2023-02-07 | 辽宁工程技术大学 | Image pair efficient dense matching method facing repeated textures and non-rigid deformation |
CN112734652A (en) * | 2020-12-22 | 2021-04-30 | 同济大学 | Near-infrared blood vessel image projection correction method based on binocular vision |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106803267B (en) | Kinect-based indoor scene three-dimensional reconstruction method | |
CN102902355B (en) | The space interaction method of mobile device | |
CN104504671B (en) | Method for generating virtual-real fusion image for stereo display | |
CN101996407B (en) | Colour calibration method for multiple cameras | |
CN104376594B (en) | Three-dimensional face modeling method and device | |
Rossi et al. | On the use of stereovision to develop a novel instrumentation system to extract geometric fire fronts characteristics | |
CN102592275B (en) | Virtual viewpoint rendering method | |
CN105184856A (en) | Two-phase human skin three-dimensional reconstruction method based on density matching | |
Tian et al. | Handling occlusions in augmented reality based on 3D reconstruction method | |
CN106780592A (en) | Kinect depth reconstruction algorithms based on camera motion and image light and shade | |
CN104574432B (en) | Three-dimensional face reconstruction method and three-dimensional face reconstruction system for automatic multi-view-angle face auto-shooting image | |
CN104376552A (en) | Virtual-real registering algorithm of 3D model and two-dimensional image | |
CN106355570A (en) | Binocular stereoscopic vision matching method combining depth characteristics | |
CN109658444B (en) | Regular three-dimensional color point cloud registration method based on multi-modal features | |
Zhao et al. | 3-D reconstruction of human body shape from a single commodity depth camera | |
KR20170073623A (en) | Fast 3d model fitting and anthropometrics | |
CN102930602B (en) | Tomography-image-based facial skin three-dimensional surface model reconstructing method | |
CN103106688A (en) | Indoor three-dimensional scene rebuilding method based on double-layer rectification method | |
CN109215085A (en) | A kind of article statistic algorithm using computer vision and image recognition | |
CN110751730B (en) | Dressing human body shape estimation method based on deep neural network | |
CN103826032A (en) | Depth map post-processing method | |
Alexiadis et al. | Fast and smooth 3d reconstruction using multiple rgb-depth sensors | |
CN109345570B (en) | Multi-channel three-dimensional color point cloud registration method based on geometric shape | |
Zhan et al. | Real-time 3D face modeling based on 3D face imaging | |
CN103307979B (en) | Based on the fruit volume measuring method of computer vision |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20151223 |
|
WD01 | Invention patent application deemed withdrawn after publication |