CN105574875B - A kind of fish eye images dense stereo matching process based on polar geometry - Google Patents
A kind of fish eye images dense stereo matching process based on polar geometry Download PDFInfo
- Publication number
- CN105574875B CN105574875B CN201510956088.9A CN201510956088A CN105574875B CN 105574875 B CN105574875 B CN 105574875B CN 201510956088 A CN201510956088 A CN 201510956088A CN 105574875 B CN105574875 B CN 105574875B
- Authority
- CN
- China
- Prior art keywords
- polar
- eye images
- fish eye
- initial
- point
- 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.)
- Active
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
A kind of fish eye images dense stereo algorithm based on polar geometry, derives the polar analytic expression of flake stero according to fisheye camera imaging model first；Then according to the searching route of specification corresponding points based on polar analytic expression and the correct window neighborhood of selection to calculate initial matching cost；Finally initial matching cost is optimized using dynamic programming algorithm to obtain the initial parallax figure of fish eye images.The inventive algorithm specification search range of corresponding points has many advantages, such as that matching speed is fast, fish eye images matching precisions is high.
Description
Technical field
The present invention relates to technical field of image processing, especially a kind of fish eye images dense stereo based on polar geometry
Method.
Background technique
Fish eye lens is a kind of shorter extreme wide-angle lens of focal length, and visual angle is close or equal to 180 degree, one figure
As can disposably capture the scenery in the even broader visual field of entire hemisphere.And the big view field imaging of more cameras composition
Although system also has above-mentioned function, high, the omnibearing imaging system based on rotary head that there are image procossing complexities
The problems such as imaging delay.Therefore, the big view field imaging system of compared to more cameras composition, fish eye images processing have simple reality
The advantages that Shi Xingqiang, is all used widely in the field for much needing big field angle.Such as: urban environment three-dimensional reconstruction, reality
When road condition monitoring, robot navigation, biped robot's step planning etc. fields.
Currently, according to the difference of obtained disparity map, the Stereo matching of fish eye images can be divided into sparse matching, quasi- dense
Matching and dense matching.Pass through the calculation by fish eye images correction at recycling conventional pinhole image after the parallel normal image of polar curve
Method completes the processing to fish eye images, and such methods thinking is simple, is relatively common fish eye images processing method.But problem is
Fish eye images deformation is serious, corrects itself is a complicated problem, and iterative processing be easy to cause details in correcting process
The loss of information, influences the effect of subsequent processing, and there is also the matching precisions low, matching of existing fish eye images matching process
The disadvantages of speed is slow.In conclusion existing fish eye images processing technique is unable to satisfy the demand of people.
Summary of the invention
That it is an object of that present invention to provide a kind of matching speeds is fast, matching precision is high, noise resisting ability is strong based on polar
The fish eye images dense stereo matching process of geometry.
To achieve the above object, use following technical scheme: the step of the method for the invention, is as follows:
Step 1, the polar analytic expression of flake stero is derived according to fisheye camera imaging model；
Step 2, according to the searching route of specification corresponding points based on polar analytic expression and the correct window neighborhood of selection
To calculate initial matching cost；
Using the search range of polar specification corresponding points, if entire image is divided into stem portion according to polar；With
Initial matching cost is calculated based on region chooses one respectively on a certain number of adjacent polar curves centered on point to be matched
The identical point of the β of fixed number amount is used as window neighborhood, while gradually clicking in the same way to each candidate on another piece image
Corresponding neighborhood is taken, initial matching cost is calculated；Solid matching method of the matching process based on region, utilizes match window inner region
The degree of correlation of grayscale information；Wherein, β is angle formed by incident ray and camera x-axis；
Step 3, initial matching cost is optimized using dynamic programming algorithm to obtain the initial parallax figure of fish eye images.
Further, in step 1, distortion first is added to fisheye camera model to optimize, then to the camera shooting of optimization
Machine model is parsed, and polar curve equation is obtained.
Further, in step 3, candidate of the point to be matched on another piece image is determined according to the polar derived
Match point carries out the calculating of initial matching cost to candidate matches point respectively using neighborhood, then using dynamic programming algorithm to first
Beginning matching cost carries out global optimization and obtains initial parallax figure.
Compared with prior art, the method for the present invention, which has the advantages that, directly carries out matching treatment to fish eye images, not
It is related to correcting process；And the matching precision of fish eye images can be effectively improved, especially to the serious image border portion that distorts
The matched precision divided, image procossing is quick, stablizes, is reliable.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is the left view of embodiment sets of views.
Fig. 3 is the right view of embodiment sets of views.
Fig. 4 is embodiment left view polar curve segmentation result figure.
Fig. 5 is embodiment right view polar curve segmentation result figure.
Fig. 6 is the disparity map being applied in embodiment sets of views using traditional rectangular window neighborhood.
Fig. 7 is to be applied to embodiment using the method for the present invention to attempt the disparity map in group.
Fig. 8 is fisheye camera projection model.
Fig. 9 is parallel binocular fisheye camera model.
Specific embodiment
The present invention will be further described with reference to the accompanying drawing:
As shown in Figure 1, the step of algorithm of the present invention, is as follows:
Step 1, the polar analytic expression of flake stero is derived according to fisheye camera imaging model；
Distortion first is added to fisheye camera model to optimize, then the camera model of optimization is parsed, is obtained
Polar curve equation.
Step 2, according to the searching route of specification corresponding points based on polar analytic expression and the correct window neighborhood of selection
To calculate initial matching cost；
Using the search range of polar specification corresponding points, if entire image is divided into stem portion according to polar；With
Region is to choose a certain number of β phases respectively on adjacent polar curve centered on point to be matched based on calculating initial matching cost
Same point gradually chooses corresponding neighborhood to each candidate point in the same way as window neighborhood, while on another piece image,
Calculate initial matching cost；Matching process utilizes the phase of match window inner region grayscale information based on the Stereo Matching Algorithm in region
Pass degree；Wherein, β is angle formed by incident ray and camera x-axis.
Step 3, initial matching cost is optimized using dynamic programming algorithm to obtain the initial parallax figure of fish eye images.
Candidate matches point of the point to be matched on another piece image is determined according to the polar derived, is distinguished using neighborhood
The calculating of initial matching cost is carried out to candidate matches point, then initial matching cost is carried out using dynamic programming algorithm global excellent
Change obtains initial parallax figure.
It is described further below using Fig. 2 and Fig. 3 as embodiment:
Fig. 2 and Fig. 3 is respectively left view and right view in sets of views.
Fig. 4 and Fig. 5 is respectively to divide the left view and right view attempted in group with polar curve, it can be seen that a left side from Fig. 4,5
Corresponding points in view and right view illustrate that the present invention is divided with identical polar curve method all on corresponding same polar curve
Left and right view.
Fig. 6 is the disparity map obtained using traditional rectangular window neighborhood method, and Fig. 7 is to be obtained using the method in the present invention
Disparity map.Fig. 6 and Fig. 7 are compared, can intuitively be obtained, the disparity map accuracy obtained using rectangular window neighborhood processing
It is low, there is the striped of similar divergence curve, and adopt the disparity map being obtained by the present invention and obtained relatively good effect, especially
Its matching precision in serious marginal portion of distorting is high, stablizes, reliably.To find out its cause, being because using conventional rectangle
When with window selection neighborhood, although the neighborhood in two windows of fish eye images central part or so can keep substantially corresponding,
But there is big distortion in the boundary part image for having arrived entire image, the neighborhood in rectangular window is no longer corresponding to be easily caused accidentally
Match.And window neighborhood choice method proposed in this paper is used, it ensure that the neighbour that left and right match window is chosen every time in entire image
The all corresponding identical scene in domain, thus the correctness of the matching result guaranteed.
Fig. 8 is fisheye camera projection model, is the model for realizing optimization after fisheye camera introducing distorts, can obtain from figure
The mapping relations of pixel m (x, y) of the point on the linear projection point p and fish eye images on camera lens spherical surface into space:
Wherein,
In formula, r is that pixel m (x, y) arrives picture centre O (u on fish eye images0,v0) distance；△ t, Δ r are to excellent
Change the distortion term of fisheye camera projection model, Δ t is the tangential distortion of fisheye camera, and Δ r is the radial distortion of fisheye camera；It is incident ray in the projection on perspective plane and the angle of x-axis.i1、i2、j1、j2、m1、m2Represent the distortion parameter of fisheye camera.
Fig. 9 is parallel binocular fisheye camera model.The optical axis of two fisheye cameras is parallel, four poles of hemispherical camera lens
Point-blank, by the analysis to model in Fig. 9, the final equation of polar curve is obtained are as follows:
In formula, i1、i2、j1、j2、m1、m2Respectively represent the distortion parameter of fisheye camera.F is the focal length of camera lens, and θ is incident
The angle of light and camera lens optical axis；α is plane (Ol, P, Or) with right camera coordinates system in oy axis institute angulation α ∈ (0, π)；β be into
Penetrate angle beta ∈ (0, π) alleged by light and right camera coordinates x-axis；It is i-th point on j-th strip polar curve and polar curve that j and i, which is respectively represented,.
It is to regard fish eye images as to be made of the individual polar curve of a rule in the present invention when finding window neighborhood,
The projection of space rectangle on the image is made of the consecutive points on several adjacent polar curves, has both avoided the parallax in match window
It discontinuously will cause error hiding, also avoid that the neighborhood in the window of parallax smooth change caused by the deformation of image is not corresponding to be made
At error hiding.
The present invention determines candidate matches point of the point to be matched on another piece image using the polar derived, right respectively
Candidate matches point carries out the calculating of initial matching cost, carries out global optimization to initial matching cost using dynamic programming algorithm and obtains
Initial parallax figure.
Matching cost formula on the neighborhood are as follows:
CSAD(xj,i,yj,i, d) and when indicating that parallax is d, i-th of point (x to be matched in right image on j-th strip polar curvej,i,
yj,i) parallax initial matching cost, (x when beinga,b',ya,b') and (xa,b,ya,b) respectively indicate neighborhood in the match window of left and right
Point.The size of neighborhood can be controlled by assigning the different value of w and h.
Embodiment described above only describe the preferred embodiments of the invention, not to model of the invention
It encloses and is defined, without departing from the spirit of the design of the present invention, those of ordinary skill in the art are to technical side of the invention
The various changes and improvements that case is made should all be fallen into the protection scope that claims of the present invention determines.
Claims (3)
1. a kind of fish eye images dense stereo matching process based on polar geometry, which is characterized in that the step of the method
It is as follows:
Step 1, the polar analytic expression of flake stero is derived according to fisheye camera imaging model；
Step 2, according to the searching route of specification corresponding points based on polar analytic expression and the correct window neighborhood of selection in terms of
Calculate initial matching cost；
Using the search range of polar specification corresponding points, if entire image is divided into stem portion according to polar；With region
Based on calculate initial matching cost choose a fixed number respectively on a certain number of adjacent polar curves centered on point to be matched
The identical point of the β of amount is used as window neighborhood, while in the same way gradually to the selection pair of each candidate point on another piece image
Neighborhood is answered, initial matching cost is calculated；Solid matching method of the matching process based on region utilizes match window inner region gray scale
The degree of correlation of information；Wherein, β is angle formed by incident ray and camera x-axis；
Step 3, initial matching cost is optimized using dynamic programming algorithm to obtain the initial parallax figure of fish eye images.
2. a kind of fish eye images dense stereo matching process based on polar geometry according to claim 1, feature
It is: in step 1, distortion first is added to fisheye camera model and is optimized, then the camera model of optimization is solved
Analysis, obtains polar curve equation.
3. a kind of fish eye images dense stereo matching process based on polar geometry according to claim 1, feature
It is: in step 3, candidate matches point of the point to be matched on another piece image is determined according to the polar derived, utilizes
Neighborhood respectively to candidate matches point carry out the calculating of initial matching cost, then using dynamic programming algorithm to initial matching cost into
Row global optimization obtains initial parallax figure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510956088.9A CN105574875B (en) | 2015-12-18 | 2015-12-18 | A kind of fish eye images dense stereo matching process based on polar geometry |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510956088.9A CN105574875B (en) | 2015-12-18 | 2015-12-18 | A kind of fish eye images dense stereo matching process based on polar geometry |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105574875A CN105574875A (en) | 2016-05-11 |
| CN105574875B true CN105574875B (en) | 2019-02-01 |
Family
ID=55884966
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510956088.9A Active CN105574875B (en) | 2015-12-18 | 2015-12-18 | A kind of fish eye images dense stereo matching process based on polar geometry |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105574875B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108074250B (en) * | 2016-11-10 | 2022-01-04 | 株式会社理光 | Matching cost calculation method and device |
| CN107909543A (en) * | 2017-12-27 | 2018-04-13 | 秦皇岛视听机械研究所 | A kind of flake binocular vision Stereo matching space-location method |
| CN110097496B (en) * | 2019-04-28 | 2020-09-01 | 燕山大学 | Fisheye image matching method based on local stable region |
| CN113436269B (en) * | 2021-06-15 | 2023-06-30 | 影石创新科技股份有限公司 | Image dense stereo matching method, device and computer equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100591608B1 (en) * | 2005-03-18 | 2006-06-20 | 전남대학교산학협력단 | Method for searching matching point in image matching |
| CN101976455A (en) * | 2010-10-08 | 2011-02-16 | 东南大学 | Color image three-dimensional reconstruction method based on three-dimensional matching |
| CN104867133A (en) * | 2015-04-30 | 2015-08-26 | 燕山大学 | Quick stepped stereo matching method |
-
2015
- 2015-12-18 CN CN201510956088.9A patent/CN105574875B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100591608B1 (en) * | 2005-03-18 | 2006-06-20 | 전남대학교산학협력단 | Method for searching matching point in image matching |
| CN101976455A (en) * | 2010-10-08 | 2011-02-16 | 东南大学 | Color image three-dimensional reconstruction method based on three-dimensional matching |
| CN104867133A (en) * | 2015-04-30 | 2015-08-26 | 燕山大学 | Quick stepped stereo matching method |
Non-Patent Citations (1)
| Title |
|---|
| 双目视觉立体匹配算法在不同介质中的研究;郝凯;《中国优秀硕士学位论文全文数据库》;20150105(第1期);第12-17页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105574875A (en) | 2016-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109345588B (en) | Tag-based six-degree-of-freedom attitude estimation method | |
| CN110569704B (en) | Multi-strategy self-adaptive lane line detection method based on stereoscopic vision | |
| CN106548173B (en) | A kind of improvement no-manned plane three-dimensional information acquisition method based on classification matching strategy | |
| CN109003311B (en) | Calibration method of fisheye lens | |
| CN105894499B (en) | A kind of space object three-dimensional information rapid detection method based on binocular vision | |
| CN110853075B (en) | Visual tracking positioning method based on dense point cloud and synthetic view | |
| CN104680496B (en) | A kind of Kinect depth map restorative procedures based on color images | |
| CN110807809B (en) | Light-weight monocular vision positioning method based on point-line characteristics and depth filter | |
| US9613420B2 (en) | Method for locating a camera and for 3D reconstruction in a partially known environment | |
| CN113362247B (en) | Semantic real scene three-dimensional reconstruction method and system for laser fusion multi-view camera | |
| CN105574875B (en) | A kind of fish eye images dense stereo matching process based on polar geometry | |
| CN106709865B (en) | Depth image synthesis method and device | |
| WO2016181687A1 (en) | Image processing device, image processing method and program | |
| CN107590827A (en) | A kind of indoor mobile robot vision SLAM methods based on Kinect | |
| CN104685513A (en) | Feature based high resolution motion estimation from low resolution images captured using an array source | |
| CN102884397A (en) | Structured-light measuring method and system | |
| CN112215880B (en) | Image depth estimation method and device, electronic equipment and storage medium | |
| CN110110793B (en) | Binocular image rapid target detection method based on double-current convolutional neural network | |
| CN113744315B (en) | Semi-direct vision odometer based on binocular vision | |
| CN104182968A (en) | Method for segmenting fuzzy moving targets by wide-baseline multi-array optical detection system | |
| CN112767546B (en) | Binocular image-based visual map generation method for mobile robot | |
| CN105513074B (en) | A kind of scaling method of shuttlecock robot camera and vehicle body to world coordinate system | |
| Ramirez et al. | Open challenges in deep stereo: the booster dataset | |
| CN110428461B (en) | Monocular SLAM method and device combined with deep learning | |
| CN105335959B (en) | Imaging device quick focusing method and its equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |