CN107330436A - A kind of panoramic picture SIFT optimization methods based on dimensional criteria - Google Patents
A kind of panoramic picture SIFT optimization methods based on dimensional criteria Download PDFInfo
- Publication number
- CN107330436A CN107330436A CN201710443220.5A CN201710443220A CN107330436A CN 107330436 A CN107330436 A CN 107330436A CN 201710443220 A CN201710443220 A CN 201710443220A CN 107330436 A CN107330436 A CN 107330436A
- Authority
- CN
- China
- Prior art keywords
- sift
- matching
- panoramic picture
- panoramic
- criterion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention proposes a kind of panoramic picture SIFT optimization methods based on dimensional criteria.The problem of for there is error hiding in SIFT matching algorithms, having invented a kind of recognize computer self and eliminates the mechanism of error hiding feature pair in omni-directional image.For the matching pair of each SIFT feature, the algorithm carries out omnidirectional imaging system criterion and dimensional criteria respectively, if two kinds of criterion conclusions have conflict, the matching makees removal processing to being considered as error hiding pair；If conflict is not present in two kinds of criterion conclusions, matches to being considered as correct matching pair, retain matching pair.Compared to traditional SIFT algorithms, the algorithm has the ability for voluntarily detecting error hiding pair, improves the matching precision of SIFT algorithms.
Description
Technical field
The invention belongs to the images match field in computer vision, and in particular to a kind of panorama sketch based on dimensional criteria
As SIFT optimization methods.
Background technology
At present, one of the problem of images match is most difficult in computer vision.Such as when space three-dimensional scene is projected as
During one secondary two dimensional image, image of the same scenery under different points of view has very big difference, and the factors in scene
Such as illumination condition, scene geometry and physical characteristic, noise jamming and distortion and camera properties, all to a certain degree
On change the gray value of image.
Images match is increasingly widely applied as the development of computer has now, and such as image matching technology can be with
Applied to military field, such as in field environment, it would be desirable to by obtained infrared view and visible images progress
Match somebody with somebody and merge, two kinds of images are combined and obtain more accurately result.Images match can also be applied to weather forecast and boat
Empty field, by correcting the change between multi-source Remote Sensing Images, realizes image co-registration, so as to obtain more fully terrestrial object information.
Image matching method is divided into by the images match based on gray scale and the images match of feature based according to matching process,
Wherein characteristic matching refers to the feature by extracting two or more images respectively, carries out parameter description to feature, then utilizes
A kind of algorithm that certain similarity measurement is matched described parameter.Scale invariant features transform method (SIFT) belongs to
The images match of feature based.
SIFT main thought is exactly the matching for changing into the matching between image between characteristic vector.Method is carried first
Stable feature to be matched is got, and is described, then the characteristic vector generated is matched.Stable treats
The feature of matching refers to that the change of image can be kept the feature of certain consistency, and ought have object of which movement, block and make an uproar
Sound shadow is maintained to the feature of preferable can match when ringing.Generally speaking, SIFT is exactly that a kind of extraction is local special
Levy, extreme point is found in metric space, extract position, yardstick, the algorithm of rotational invariants, SIFT is that a kind of robustness is good, chi
The constant character description method of degree, is obtained extensively in fields such as image registration, image mosaic, household articles classification and recognitions of face
General application, SIFT shortcoming is also a lot, and such as time complexity, needs big compared with long, calculating data volume time-consuming compared with high, algorithm are artificial dry
It is pre- etc..
SIFT shortcoming is directed to, domestic and foreign scholars are being directed to optimizing this algorithm, such as patent (publication number in recent years always
For CN104834931A) a kind of improved scale invariant feature matching algorithm based on wavelet transformation is proposed, original classical
Work algorithm on the basis of, introduce two-dimentional fast wavelet transform algorithm, the low-frequency component of reconstruction image, then to gaussian pyramid
Group number is adjusted, and reduces down-sampled number of times, and Mismatching point is rejected finally by the algorithm of optimization.Algorithm after improvement not only subtracts
Lack matching time-consuming, and matching rate is also improved.But current image matching technology also face can not realize it is complete
The problem of Auto-matching, it is intended that on the premise of any manual intervention is not needed, computer can be according to the program of setting
Be automatically performed the matching between multi-source image, be this we have proposed a kind of panoramic picture SIFT optimization sides based on dimensional criteria
Method, the algorithm has the ability for voluntarily detecting error hiding pair, and improves the matching precision of SIFT algorithms.
The content of the invention
It is an object of the invention to provide a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria, compared to tradition
SIFT algorithms, this algorithm has the ability for voluntarily detecting error hiding pair, improves the matching precision of SIFT algorithms.
What the present invention was realized in：
A kind of panoramic picture SIFT optimized algorithms based on dimensional criteria, concrete implementation step is as follows：
Step 1. shoots several panorama sketch respectively using diverse location of the panoramic vision imaging system in same level
Picture；
Step 2. increases the number of plies in original SIFT algorithms and group number, and multiple image is carried out SIFT feature extract with
Matching, obtains SIFT feature matching pair；
Step 3. carries out omnidirectional imaging system criterion and dimensional criteria respectively for the matching pair of each SIFT feature；
Step 4. compares omnidirectional imaging system criterion and dimensional criteria, if two kinds of criterion knots for each SIFT matchings pair
By there is conflict, then the matching is to being considered as error hiding pair, and makees removal processing, if conflict is not present in two kinds of criterion conclusions,
Matching retains matching pair to being considered as correct matching pair；
Step 5. checks whether the matching pair not judged also, if there is the matching pair not judged, continues repeat step
3rd, 4,5, if all matchings are to having judged, program, which is performed, to be terminated.
The width number of panoramic picture described in step 1 is at least 2.
Panoramic imagery criterion described in step 3 includes panoramic imagery criterion 1 and panoramic imagery criterion 2, for each
SIFT feature matching pair, judges picture position and the position relationship of panoramic picture annulus of two characteristic points of matching centering, if
Characteristic point then performs panoramic imagery criterion 1, if characteristic point is held outside panoramic picture annulus in panoramic picture annulus
Row panoramic imagery criterion 2.
Dimensional criteria regulation described in step 3, for the matching pair of each SIFT feature, judges two of matching centering
The scale size relation of characteristic point simultaneously performs dimensional criteria, when the scale-value of SIFT feature is bigger, illustrates that camera site is arrived
Actual range representated by this feature point is smaller, when the scale-value of SIFT feature is smaller, illustrates camera site to this feature
The representative actual range of point is bigger.
In dimensional criteria described in step 3, on the premise of matching precision is ensured, increase the image layer of metric space
Number.
Described panoramic picture annulus refer to being located at optical axis natural landmark in the horizontal plane be mapped to panoramic picture
When middle, imaging point can only be located on an annulus in panoramic picture, and no matter what kind of imaging system, which carries out, moves horizontally, and is located at
The natural landmark imaging point of optical axis horizontal plane will not leave this panoramic picture annulus.
The beneficial effects of the present invention are：
The present invention is directed to panoramic picture feature extraction, devises a kind of panoramic picture SIFT optimizations based on dimensional criteria and calculates
Method.By setting panoramic picture criterion and dimensional criteria, computer can voluntarily detect and remove Mismatching point, improve spy
Levy Point matching precision.This method has carried out algorithm optimization in traditional SIFT extraction algorithms, effectively increases the algorithm degree of accuracy, should
Algorithm can be widely applied to the fields such as image procossing, pattern-recognition, robot navigation.
Brief description of the drawings
Fig. 1 is system flow chart of the invention.
Fig. 2 is the front view of omnidirectional imaging system criterion in the present invention.
Fig. 3 is the top view of omnidirectional imaging system criterion in the present invention.
Embodiment
The present invention is described further below in conjunction with the accompanying drawings.
As shown in figure 1, being the system flow chart of the present invention.Detailed process is as follows：
Step 1. shoots two width panorama sketch respectively using diverse location of the panoramic vision imaging system in same level
Picture；
Step 2. adjusts the number of plies in original SIFT algorithms to 6 by 3, and carries out SIFT feature extraction to two images
With matching；
Step 3. judges picture position and the panorama of two characteristic points of matching centering for the matching pair of each SIFT feature
The position relationship of image annulus, if characteristic point is in annulus, performs panoramic imagery criterion 1, if outside annulus, performing
Panoramic imagery criterion 2；For the matching pair of each SIFT feature, the scale size relation of two characteristic points of matching centering is judged,
And perform dimensional criteria；
Whether the conclusion of two kinds of criterions in step 4. checking procedure 3 conflicts.If conflict, match to being considered as mistake
Pairing, gives removal processing, if not conflicting, and matches to being considered as correct matching pair, is retained；
Step 5. checks whether the matching pair not judged also, if there is the matching pair not judged, continues repeat step
3rd, 4,5, if all matchings are to having judged, program, which is performed, to be terminated.
It is the front view and top view of omnidirectional imaging system criterion in the present invention, wherein Fig. 2 is just as shown in Fig. 2 and Fig. 3
View, Fig. 3 is top view.According to the image-forming principle of panoramic vision imaging system, most environmental information all may be used in space
To be mapped on a Zhang Quanjing image.A' in Fig. 2, B', C', D', H1' and H2' it is 6 road signs in real scene, wherein H1'
With H2' positioned at the horizontal plane where the optical axis of panoramic vision imaging system, A' and B' is respectively positioned on above optical axis and vertical height phase
Together, C' is respectively positioned on below optical axis with D' and vertical height is identical, F1With F2It is two focuses of panoramic vision imaging system, in order to
The connected mode between road sign and imaging point corresponding with road sign in Fig. 2 is distinguished, the straight line color that regulation is connected between A' and A is
Straight line color between black, connection B' and B is that the straight line color between Dark grey, connection D' and D is light gray.A in Fig. 3,
B,C,D,H1With H2It is road sign A', B', C', D', H1' and H2' imaging point in each comfortable panoramic picture.
From Fig. 2 and Fig. 3, positioned at optical axis natural landmark in the horizontal plane be mapped in panoramic picture when, imaging
Point can only be located on an annulus in panoramic picture, and no matter what kind of imaging system, which carries out, moves horizontally, positioned at optical axis level
The natural landmark imaging point in face will not leave annulus；And it is located at natural landmark more than optical axis, corresponding imaging point is respectively positioned on
Outside annulus, and on the premise of vertical height is constant, the distance of road sign and camera site is more remote, then imaging point is closer to annulus,
Close to picture centre；The corresponding imaging point of natural landmark below optical axis is respectively positioned in annulus, and constant in vertical height
On the premise of, the distance of road sign and camera site is more remote, then imaging point is closer to annulus, away from picture centre.
According to above phenomenon, figure of two SIFT features of the invention by judging characteristic centering to respective picture centre
Image distance is from size, to judge corresponding actual road sign to the relative distance of two camera sites.If { fP,fQIt is in P position and Q
A pair of characteristic matchings pair in the two width panoramic pictures that position is shot；L is { fP,fQRepresentated by real scene in natural road
Mark；OIPWith OIQIt is the center of two images.Therefore by calculating image distance d (fP,OIP) and d (fQ,OIQ) relation, come
Judge actual range d (L, P) and d (L, Q) relation, therefore the omnidirectional imaging system criterion that the present invention is utilized is as follows：
Work as fPWith fQWhen being respectively positioned on inside panoramic picture annulus, panoramic imagery criterion 1 is denoted as, expression formula is：
Work as fPWith fQWhen being respectively positioned on outside panoramic picture annulus, panoramic imagery criterion 2 is denoted as, expression formula is：
Dimensional information by SIFT algorithm mesoscales spatial extrema detection with key point location and obtain, SIFT passes through
By initial picture carry out continuous Gaussian Blur with it is down-sampled and form multigroup, multilayer conversion picture so that it is empty to form yardstick
Between, and each SIFT points are then produced from metric space, therefore each SIFT points are respectively provided with its distinctive scale-value.And
Scale-value can provide an important information：If some natural landmark in real scene is nearer apart from camera site, past
It is past to need the higher Gaussian smoothing of fog-level, that is to say, that its corresponding SIFT feature often has bigger yardstick
Value.Therefore, for above phenomenon, the present invention increases the image number of plies of metric space first, so that before matching precision is ensured
Put, improve SIFT feature Point matching sum, enrich dimensional information；And by judging matching to { fP,fQIn two features
The scale-value σ of pointPWith σQSize, to judge this feature to the distance to two camera sites.As dimensional criteria, as follows：
As the judgment criteria of two different angles, omnidirectional imaging system criterion may determine that nature road with dimensional criteria
Mark the relative distance relation of actual photographed position.Therefore for the matching pair of some SIFT feature, if obtained by the two criterions
The conclusion gone out is collided with each other, then illustrate the matching to essential information necessarily disagreed with one of criterion, therefore can demonstrate,prove
The bright matching is to being necessarily error hiding pair.
Claims (6)
1. a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria, it is characterised in that concrete implementation step is as follows：
Step 1. shoots several panoramic pictures respectively using diverse location of the panoramic vision imaging system in same level；
Step 2. increases the number of plies in original SIFT algorithms and group number, and multiple image is carried out SIFT feature extract with
Match somebody with somebody, obtain SIFT feature matching pair；
Step 3. carries out omnidirectional imaging system criterion and dimensional criteria respectively for the matching pair of each SIFT feature；
Step 4. compares omnidirectional imaging system criterion and dimensional criteria, if two kinds of criterion conclusions are deposited for each SIFT matchings pair
In conflict, then the matching is to being considered as error hiding pair, and makees removal processing, if conflict is not present in two kinds of criterion conclusions, matches
To being considered as correct matching pair, retain matching pair；
Step 5. checks whether the matching pair not judged also, if there is the matching pair that does not judge, continue repeat step 3,4,
5, if all matchings are to having judged, program, which is performed, to be terminated.
2. a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria according to claim 1, it is characterised in that：Step
The width number of panoramic picture described in rapid 1 is at least 2.
3. a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria according to claim 1, it is characterised in that：Step
Panoramic imagery criterion described in rapid 3 includes panoramic imagery criterion 1 and panoramic imagery criterion 2, is matched for each SIFT feature
It is right, picture position and the position relationship of panoramic picture annulus of two characteristic points of matching centering are judged, if characteristic point is complete
In scape image annulus, then panoramic imagery criterion 1 is performed, if characteristic point is outside panoramic picture annulus, performs panoramic imagery and sentence
According to 2.
4. a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria according to claim 1, it is characterised in that：Step
Dimensional criteria regulation described in rapid 3, for the matching pair of each SIFT feature, judges the chi of two characteristic points of matching centering
Degree magnitude relationship simultaneously performs dimensional criteria, when the scale-value of SIFT feature is bigger, illustrates camera site to this feature point institute
The actual range of representative is smaller, when the scale-value of SIFT feature is smaller, illustrates representated by camera site to this feature point
Actual range is bigger.
5. a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria according to claim 1, it is characterised in that：Step
In dimensional criteria described in rapid 3, on the premise of matching precision is ensured, increase the image number of plies of metric space.
6. a kind of panoramic picture SIFT optimized algorithms based on dimensional criteria according to claim 3, it is characterised in that：Institute
The panoramic picture annulus stated refer to being located at optical axis natural landmark in the horizontal plane when being mapped in panoramic picture, imaging point
It can only be located on an annulus in panoramic picture, no matter what kind of imaging system, which carries out, moves horizontally, positioned at optical axis horizontal plane
Natural landmark imaging point will not leave this panoramic picture annulus.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710443220.5A CN107330436B (en) | 2017-06-13 | 2017-06-13 | Scale criterion-based panoramic image SIFT optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710443220.5A CN107330436B (en) | 2017-06-13 | 2017-06-13 | Scale criterion-based panoramic image SIFT optimization method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107330436A true CN107330436A (en) | 2017-11-07 |
| CN107330436B CN107330436B (en) | 2020-07-28 |
Family
ID=60195554
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710443220.5A Active CN107330436B (en) | 2017-06-13 | 2017-06-13 | Scale criterion-based panoramic image SIFT optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107330436B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805799A (en) * | 2018-04-20 | 2018-11-13 | 平安科技（深圳）有限公司 | Panoramic picture synthesizer, method and computer readable storage medium |
| CN110245566A (en) * | 2019-05-16 | 2019-09-17 | 西安交通大学 | A kind of remote method for tracing of infrared target based on background characteristics |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080106594A1 (en) * | 2006-11-07 | 2008-05-08 | The Board Of Trustees Of The Leland Stanford Jr. University | System and method for tagging objects in a panoramic video and associating functions and indexing panoramic images with same |
| CN101833765A (en) * | 2010-04-30 | 2010-09-15 | 天津大学 | Characteristic matching method based on bilateral matching and trilateral restraining |
| EP2237227A1 (en) * | 2009-04-01 | 2010-10-06 | Nederlandse Organisatie voor toegepast -natuurwetenschappelijk onderzoek TNO | Video sequence processing method and system |
| CN106355555A (en) * | 2016-10-24 | 2017-01-25 | 北京林业大学 | Image stitching method and device |
| CN106530407A (en) * | 2016-12-14 | 2017-03-22 | 深圳市金大象文化发展有限公司 | Three-dimensional panoramic splicing method, device and system for virtual reality |
-
2017
- 2017-06-13 CN CN201710443220.5A patent/CN107330436B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080106594A1 (en) * | 2006-11-07 | 2008-05-08 | The Board Of Trustees Of The Leland Stanford Jr. University | System and method for tagging objects in a panoramic video and associating functions and indexing panoramic images with same |
| EP2237227A1 (en) * | 2009-04-01 | 2010-10-06 | Nederlandse Organisatie voor toegepast -natuurwetenschappelijk onderzoek TNO | Video sequence processing method and system |
| CN101833765A (en) * | 2010-04-30 | 2010-09-15 | 天津大学 | Characteristic matching method based on bilateral matching and trilateral restraining |
| CN106355555A (en) * | 2016-10-24 | 2017-01-25 | 北京林业大学 | Image stitching method and device |
| CN106530407A (en) * | 2016-12-14 | 2017-03-22 | 深圳市金大象文化发展有限公司 | Three-dimensional panoramic splicing method, device and system for virtual reality |
Non-Patent Citations (1)
| Title |
|---|
| QIDAN ZHU ET AL.: "A Novel Robot Visual Homing Method Based on SIFT Features", 《SENSORS》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805799A (en) * | 2018-04-20 | 2018-11-13 | 平安科技（深圳）有限公司 | Panoramic picture synthesizer, method and computer readable storage medium |
| CN108805799B (en) * | 2018-04-20 | 2021-04-23 | 平安科技（深圳）有限公司 | Panoramic image synthesis apparatus, panoramic image synthesis method, and computer-readable storage medium |
| CN110245566A (en) * | 2019-05-16 | 2019-09-17 | 西安交通大学 | A kind of remote method for tracing of infrared target based on background characteristics |
| CN110245566B (en) * | 2019-05-16 | 2021-07-13 | 西安交通大学 | Infrared target remote tracking method based on background features |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107330436B (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111126412B (en) | Image key point detection method based on characteristic pyramid network | |
| US8467596B2 (en) | Method and apparatus for object pose estimation | |
| CN104850850B (en) | A kind of binocular stereo vision image characteristic extracting method of combination shape and color | |
| CN107103589B (en) | A kind of highlight area restorative procedure based on light field image | |
| CN110097093A (en) | A kind of heterologous accurate matching of image method | |
| CN104599258B (en) | A kind of image split-joint method based on anisotropic character descriptor | |
| CN104751465A (en) | ORB (oriented brief) image feature registration method based on LK (Lucas-Kanade) optical flow constraint | |
| CN103902953B (en) | A kind of screen detecting system and method | |
| CN102834845A (en) | Method and arrangement for multi-camera calibration | |
| CN111986240A (en) | Drowning person detection method and system based on visible light and thermal imaging data fusion | |
| CN113159043B (en) | Feature point matching method and system based on semantic information | |
| CN104182973A (en) | Image copying and pasting detection method based on circular description operator CSIFT (Colored scale invariant feature transform) | |
| CN106643555B (en) | Connector recognition methods based on structured light three-dimensional measurement system | |
| CN111192194B (en) | Panoramic image stitching method for curtain wall building facade | |
| Zhang et al. | License plate localization in unconstrained scenes using a two-stage CNN-RNN | |
| CN110378995B (en) | Method for three-dimensional space modeling by using projection characteristics | |
| CN117036641A (en) | Road scene three-dimensional reconstruction and defect detection method based on binocular vision | |
| CN109086350B (en) | Mixed image retrieval method based on WiFi | |
| CN107154017A (en) | A kind of image split-joint method based on SIFT feature Point matching | |
| CN108647580A (en) | Improved SIFT-based ISAR image feature point extraction and matching method | |
| CN110110793A (en) | Binocular image fast target detection method based on double-current convolutional neural networks | |
| CN105469042A (en) | Improved face image comparison method | |
| JP5367244B2 (en) | Target detection apparatus and target detection method | |
| CN107330436A (en) | A kind of panoramic picture SIFT optimization methods based on dimensional criteria | |
| Liu et al. | Keypoint matching by outlier pruning with consensus constraint |
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 |