CN106971381A - A kind of wide angle camera visual field line of demarcation generation method with the overlapping ken - Google Patents
A kind of wide angle camera visual field line of demarcation generation method with the overlapping ken Download PDFInfo
- Publication number
- CN106971381A CN106971381A CN201710150152.3A CN201710150152A CN106971381A CN 106971381 A CN106971381 A CN 106971381A CN 201710150152 A CN201710150152 A CN 201710150152A CN 106971381 A CN106971381 A CN 106971381A
- Authority
- CN
- China
- Prior art keywords
- image
- demarcation
- visual field
- field line
- 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.)
- Granted
Links
- 230000000007 visual effect Effects 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000012937 correction Methods 0.000 claims abstract description 36
- 238000013507 mapping Methods 0.000 claims abstract description 15
- 238000005457 optimization Methods 0.000 claims description 2
- 238000013499 data model Methods 0.000 claims 1
- 239000011159 matrix material Substances 0.000 claims 1
- 230000009466 transformation Effects 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 5
- 238000012935 Averaging Methods 0.000 abstract 1
- 238000001514 detection method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000005603 centrifugal distortion Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Studio Devices (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
A kind of visual field line of demarcation generation method for being directed to the wide angle camera with the overlapping ken.By the way of back projection, the visual field line of demarcation of correcting image is projected back under wide angle picture.Comprise the following steps:Pass through Zhang Zhengyou standardizations and multinomial distortion correction model first, fault image is corrected, by extracting SIFT feature and deleting error hiding by RANSAC algorithms, realize and image registration is carried out to adjacent two images, by generating visual field line of demarcation under the image of the method for averaging after calibration, finally by image pixel coordinates corresponding relation formula before and after distortion correction, visual field line of demarcation is projected back in wide-angle figure.This method introduces distortion for distortion camera causes traditional images analysis method to be unable to application, using the method for back mapping, the accurate visual field line of demarcation obtained under fault image.Target handoff the wide angle cameras for having overlapping region to realization, there is significant application value.
Description
Technical field
The invention belongs to technical field of image processing, it is related to automatically generate under a kind of wide angle camera with the overlapping ken and regards
The method in wild line of demarcation.
Background technology
Video monitoring system is increasingly extensive in the application of business, national defence and military field, currently to intelligent monitor system
Research, emphasis is target detection, tracking and behavioural analysis and identification again in terms of.Target following in the visual field is most important
Key technology, is the basis of follow-up various technologies.Obtain wide visual field and realize prolonged target with lasting track, it is necessary to
Expanded field of vision is come by using camera combination.The information exchange between different cameras is generated the problem that therewith with merging, it is right
Target following under camera combination, seeks to whether judge target across the ken, and matches under neighbouring camera
Same target, realize goal congruence differentiate.Further, since having the advantages that big visual field, short focus, wide-angle lens is extended
The field range of common pinhole camera, is also frequently used in intelligent monitor system, but wide-angle lens can introduce significant abnormal
Become, this will cause conventional analysis of image content algorithm to be no longer applicable.To realize particular visual algorithm application, it is necessary to for
Wide angle camera shooting imaging characteristicses carry out method adjustment.
Visual field line of demarcation (FOV:Field OF View) refer in the overlapping multi-camera system of the ken, adjacent camera
Between public observation part boundary line.Common practice is to extract the characteristic point in two images, realizes characteristic matching point, so
Some error hidings are filtered out by the projection relation contained between match point afterwards, calculating two images correspondence by projection relation closes
System, completes the visual field line of demarcation generation of visual field lap.
However, under wide angle cameras, because being influenceed visual field line of demarcation not to be straight line but curve by distorting.At this
In the case of kind, realize that characteristic matching determines that dosing control has larger difficulty, it is impossible to efficiently generate visual field line of demarcation.
The content of the invention
Scope of application limitation for generally generating visual field boundary line method, is directed to overlapping the invention provides one kind
The visual field line of demarcation generation method of the wide angle camera of the ken.By the way of back projection, by the visual field line of demarcation of correcting image
Back projection is returned under wide angle picture.Experiment shows that this method is efficiently solved in the presence of the multiple-camera visual field point in the case of distortion
Boundary line generates the visual field under problem, the application scenario of the analysis of image content for needing multichannel wide-angle camera, wide angle camera
Line of demarcation generation method has suitable applied value.
The visual field line of demarcation generation method of the present invention, carries out distortion correction by wide angle camera first, by characteristic matching,
Visual field line of demarcation is generated under image after distortion correction, by the way of back projection, visual field line of demarcation is projected back in wide-angle figure
As the lower final line of demarcation of generation.First by Zhang Zhengyou standardizations and multinomial distortion model, fault image is corrected, led to
Cross extraction SIFT feature and adjacent two images are carried out with generation visual field boundary under image registration, the image after distortion correction
Line, then by image pixel coordinates corresponding relation formula before and after distortion correction, generates projection table, by the method tabled look-up by the visual field
Line of demarcation, which is projected back under wide angle picture, generates final line of demarcation.Comprise the following steps that:
Step one:Wide angle cameras is demarcated using Zhang Zhengyou standardizations, camera parameters and correction model is obtained
Parameter.
For wide angle cameras correction, excessively complicated modeling often carrys out counter productive to correction tape.In current big portion
Partial image handles application field, can only consider radial distortion, it becomes possible to obtain good calibration result.Ignore other distortion shapes
Formula (such as centrifugal distortion and tangential distortion), only retains radial distortion, here using simple positive multinomial calibration model, fixed
Adopted ruThe distance for any pixel coordinate point P ' for being principal point into ideal image.Similarly, rdPrincipal point is defined as into fault image
Any pixel P distance.The centre coordinate point of image, which mutually coincides, before and after distortion can obtain correcting front and rear using pixel as amount
Image coordinate relation under the image coordinate system of guiding principle:
Wherein k1, k2, k3 ... kn are the distortion correction parameter of optimization undetermined, and (u, v) is the pixel coordinate of fault image,
(u ', v ') is the pixel coordinate of image after correction, (u0,v0) it is principal point coordinate.
In general, multinomial model gets the correcting distorted wide angle picture of the 5th rank with regard to that can meet most of demands.Cause
This n value is often set to 5.So (1) formula can use following formula approximate substitution:
Whole parameters in this model are tried to achieve using Zhang Zhengyou methods.(2) model provided is schemed before and after resulting in correction
The coordinate corresponding relation of picture, completes correction purpose.
Step 2:To image zooming-out SIFT feature after correction, characteristic matching is carried out:At present on how to determine two spies
Also ununified model for accurate calculation can a little be matched by levying.In the case, according to the Euclidean distance between feature descriptor
Linked character point is regarded as a kind of fairly simple effective method.Assuming that a pair of descriptor desp,desq, then between the two
Distance is
Because SIFT feature vector dimension is very high, and there are some closely similar characteristic points, cause many non-matching
Distance value of the point pair between matching double points is close and is difficult to directly differentiation.Assume if based on such a:One characteristic point exists
The point that presence can be matched in some characteristic points, then its distance between the match point and the distance between non-matching point have aobvious
Write gap, then a feasible method is to calculate each characteristic point and so the Euclidean distance of point to be matched, finds out minimum
Distance and time minimum range, another minimum range and time minimum range carry out ratio computing, when ratio is less than setting value, show most
The corresponding point of small distance can be with Feature Points Matching.
Step 3:The matching carried out using the hypothesis in the upper section characteristic point matching method, it may appear that have to a certain degree
Mistake.It is therefore desirable to matching result to screening after carry out subsequent treatment again, using RANSAC (Random
Sample Consensus) algorithm can effectively solve in distinguished point based registration technique reject error matching points the problem of.
Step 4:Corresponding Feature Points Matching pair is finally obtained under two images, it is horizontal to the characteristic point under each image
Coordinate takes average respectively, as the abscissa in line of demarcation, passes through the visual field line of demarcation of image after the characteristic point generation correction of matching.
Step 5:By image coordinate corresponding relation before and after the correction that is obtained in step one, by the visual field line of demarcation of generation
Projected to by back mapping in original wide angle picture.
Here back mapping is realized by building the front and rear location of pixels mapping table of correction:Above-mentioned multinomial calibration model, energy
Realize that flake is corrected relatively simplely, but it cannot be guaranteed that all images coordinate after correction has corresponding pixel value.Here
Using the missing of image pixel positions after the method supplement correction of arest neighbors difference, i.e., by formula (1), calculate fault image picture
Vegetarian refreshments P (x, y) corresponds to the pixel coordinate (x ', y ') of image after correction, and the picture is recorded in mapping table (x ', y ') position of structure
Plain coordinate (x, y).Finally for blank position in form, the mode for taking consecutive value is simply taken, proximal most position is taken, apart from phase
Leftward position value is taken Deng in the case of.The whole positions of fault image are traveled through, mapping table is completed and builds.
The visual field line of demarcation position that image is generated after the back projection in visual field line of demarcation, i.e. traversal correction is realized by mapping table
Put, correspondence is obtained in the position of fault image by searching projection table.Complete the visual field line of demarcation under final adjacent fault image
Generation, recognizes as successive image, analyzes, handling and use.
The inventive method introduces distortion for distortion camera causes traditional images analysis method to be unable to application, uses
The method of back mapping, after field of view line of demarcation generation after completing correction, the accurate visual field boundary obtained under fault image
Line.Target handoff the wide angle cameras for having overlapping region to realization, there is significant application value.
Brief description of the drawings
Fig. 1 is the wide-angle distortion figure of the present invention.
Fig. 2 is the multinomial calibration result figure of the present invention.
Fig. 3 is the feature point detection and matching effect figure of the present invention.
Fig. 4 is the feature point detection screening effect figure of the present invention.
Fig. 5 is the correcting image visual field line of demarcation generation of the present invention.
Fig. 6 is the original image visual field line of demarcation generation of the present invention.
Embodiment
The specific implementation to the present invention elaborates below in conjunction with the accompanying drawings.
The method of present embodiment, software environment is the systems of WINDOWS 7, and simulated environment is MATLAB2013a.
Step one:Wide-angle camera is demarcated.150 ° of the wide-angle camera field range wherein used, resolution ratio is
640*480, obtained camera relevant parameter is as follows:
(1) principal point coordinate value:u0=267.3437, v0=362.0111
(2) distortion correction parameter ki value is as follows:
K1=0.36;K2=0.38;K3=0.0016;K4=0.00000078;K5=-1.3297*10-10
By relational expression (1), the corresponding relation of image pixel coordinates position before and after correction is obtained, is caused for image zooming
Image cavity, using the method for closest difference.Original image is as shown in figure 1, image is as shown in Figure 2 after correction.
Step 2:Two images SIFT feature after correction is extracted, for a bit in piece image, is calculated and another width
Euclidean distance in image between the feature descriptor of all characteristic points, finds out minimum range and time minimum range, carries out ratio
Computing, when ratio is less than setting value, shows that the corresponding point of minimum range can be with Feature Points Matching.All characteristic points are traveled through,
Complete characteristic matching.Matching result is as shown in Figure 3.
Step 3:It can see from the result directly matched and there are a large amount of error matching points, excluded using RANSAC algorithms
Exterior point, process is as follows
(1) one group of measurement data is randomly choosed, it is assumed that be entirely intra-office point.
(2) temporary pattern is calculated according to sample in (1).
(3) data are chosen according to test with obtained model, if enough points are classified as the intra-office point of hypothesis,
The model so estimated is just reasonable enough.
(4) then, go to reevaluate model with the intra-office point of all hypothesis, by estimate the error of intra-office point and model come
Assessment models, are classified as intra-office point and then give up the model very little, are selected if existing model is better than.
(5) repeat above-mentioned steps, until computation model number of times reaches setting number of times, or be classified as interior point probability it is big
In specified probability.
The matching result for removing erroneous matching is as shown in Figure 4.
Step 4:Matching characteristic point horizontal position coordinate average value in the result of calculation procedure three, as boundary line position, is painted
The visual field line of demarcation of image after vertical line processed is corrected.As shown in Figure 5.
Step 5:By formula one, mapping table is generated, the visual field line of demarcation that step 4 is generated is anti-by searching mapping table
To the visual field line of demarcation being mapped between original image, generation original image.Fig. 6 gives the generation effect in visual field line of demarcation.Can
To find out that, by distortion correction and back mapping, the line of demarcation of generation relatively accurately reflects the correspondence of actual scene.Can be with
Carry out subsequent processing steps.
In summary, the visual field line of demarcation generation method based on back mapping realizes that the wide angle camera with the overlapping ken is regarded
The generation in wild line of demarcation.This method compared to directly carry out feature extraction generate visual field line of demarcation method, detect it is more into
The matching double points of work(, generate more accurate visual field line of demarcation, realize target handoff.
Claims (2)
1. a kind of visual field line of demarcation generation method of the wide angle camera with the overlapping ken, it is characterised in that following steps,
Step one:Using Zhang Zhengyou calibration algorithms and multinomial distortion correction model, only consider to calculate under the conditions of radial distortion
Correction coefficient k in positive multinomial modeli, distortion correction is carried out to image;
Using positive multinomial calibration model, r is defineduThe distance for any pixel coordinate point P ' for being principal point into ideal image;Together
Reason, rdIt is defined as any pixel P of the principal point into fault image distance;The centre coordinate point of image is weighed mutually before and after distortion
Close obtain correction before and after using pixel as the image coordinate system of dimension under image coordinate relation:
Wherein k1, k2, k3 ... kn are the distortion correction parameter of optimization undetermined, and (u, v) is the pixel coordinate of fault image, (u ',
V ') for correction after image pixel coordinate, (u0,v0) it is principal point coordinate;
Step 2:To image zooming-out SIFT feature after correction, according to the Euclidean distance linked character point between feature descriptor,
Carry out characteristic matching;
Assuming that a pair of descriptor desp,desq, then distance between the two be
Calculate each characteristic point and so the Euclidean distance of point to be matched, finds out minimum range and time minimum range, another minimum
Distance and time minimum range carry out ratio computing, when ratio is less than setting value, show that the corresponding point of minimum range can be with spy
Levy Point matching;
Step 3:Using RANSAC algorithms, using projective transformation matrix as data model, the erroneous matching result of step 2 is removed;
Step 4:Corresponding Feature Points Matching pair is finally obtained under two images, to the characteristic point abscissa under each image
Average is taken respectively, as the abscissa in line of demarcation, passes through the visual field line of demarcation of image after the characteristic point generation correction of matching;
Step 6:According to image pixel coordinates corresponding relation formula before and after the correction obtained in step one, mapping table is generated, by looking into
Look for mapping table by the result back mapping in step 3 to original image, obtain final visual field line of demarcation.
2. visual field line of demarcation as claimed in claim 1 generation method, is further characterized in that, n value is 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710150152.3A CN106971381B (en) | 2017-03-14 | 2017-03-14 | A kind of wide angle camera visual field line of demarcation generation method with the overlapping ken |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710150152.3A CN106971381B (en) | 2017-03-14 | 2017-03-14 | A kind of wide angle camera visual field line of demarcation generation method with the overlapping ken |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106971381A true CN106971381A (en) | 2017-07-21 |
CN106971381B CN106971381B (en) | 2019-06-18 |
Family
ID=59329373
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710150152.3A Active CN106971381B (en) | 2017-03-14 | 2017-03-14 | A kind of wide angle camera visual field line of demarcation generation method with the overlapping ken |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106971381B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108053385A (en) * | 2018-01-24 | 2018-05-18 | 桂林电子科技大学 | A kind of real-time correction system of flake video and method |
CN110430400A (en) * | 2019-08-12 | 2019-11-08 | 中国人民解放***箭军工程大学 | A kind of ground level method for detecting area of the movable video camera of binocular |
CN116912517A (en) * | 2023-06-06 | 2023-10-20 | 阿里巴巴(中国)有限公司 | Method and device for detecting camera view field boundary |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140044325A1 (en) * | 2012-08-09 | 2014-02-13 | Hologic, Inc. | System and method of overlaying images of different modalities |
CN103997624A (en) * | 2014-05-21 | 2014-08-20 | 江苏大学 | Overlapped domain dual-camera target tracking system and method |
-
2017
- 2017-03-14 CN CN201710150152.3A patent/CN106971381B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140044325A1 (en) * | 2012-08-09 | 2014-02-13 | Hologic, Inc. | System and method of overlaying images of different modalities |
CN103997624A (en) * | 2014-05-21 | 2014-08-20 | 江苏大学 | Overlapped domain dual-camera target tracking system and method |
Non-Patent Citations (2)
Title |
---|
SENEM VELIPASALAR ET.AL: "RECOVERING FIELD OF VIEW LINES BY USING PROJECTIVE INVARIANTS", 《2004 INTERNATIONAL CONFERENCE ON IMAGE PROCESSING(ICIP)》 * |
杨俊 等: "基于SIFT及射影变换的摄像机视野分界线恢复", 《图像图形技术研究与应用》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108053385A (en) * | 2018-01-24 | 2018-05-18 | 桂林电子科技大学 | A kind of real-time correction system of flake video and method |
CN110430400A (en) * | 2019-08-12 | 2019-11-08 | 中国人民解放***箭军工程大学 | A kind of ground level method for detecting area of the movable video camera of binocular |
CN110430400B (en) * | 2019-08-12 | 2020-04-24 | 中国人民解放***箭军工程大学 | Ground plane area detection method of binocular movable camera |
CN116912517A (en) * | 2023-06-06 | 2023-10-20 | 阿里巴巴(中国)有限公司 | Method and device for detecting camera view field boundary |
CN116912517B (en) * | 2023-06-06 | 2024-04-02 | 阿里巴巴(中国)有限公司 | Method and device for detecting camera view field boundary |
Also Published As
Publication number | Publication date |
---|---|
CN106971381B (en) | 2019-06-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xue et al. | Learning to calibrate straight lines for fisheye image rectification | |
Sochor et al. | Traffic surveillance camera calibration by 3d model bounding box alignment for accurate vehicle speed measurement | |
CN107240124B (en) | Cross-lens multi-target tracking method and device based on space-time constraint | |
Boltes et al. | Automatic extraction of pedestrian trajectories from video recordings | |
CN112883819A (en) | Multi-target tracking method, device, system and computer readable storage medium | |
CN102313536B (en) | Method for barrier perception based on airborne binocular vision | |
Chang et al. | Tracking Multiple People Under Occlusion Using Multiple Cameras. | |
CN110992263B (en) | Image stitching method and system | |
TWI639136B (en) | Real-time video stitching method | |
Tang et al. | ESTHER: Joint camera self-calibration and automatic radial distortion correction from tracking of walking humans | |
CN111723801B (en) | Method and system for detecting and correcting target in fisheye camera picture | |
CN110189375B (en) | Image target identification method based on monocular vision measurement | |
CN101860729A (en) | Target tracking method for omnidirectional vision | |
CN102243765A (en) | Multi-camera-based multi-objective positioning tracking method and system | |
US8428313B2 (en) | Object image correction apparatus and method for object identification | |
CN112287867B (en) | Multi-camera human body action recognition method and device | |
CN105809626A (en) | Self-adaption light compensation video image splicing method | |
CN111160291B (en) | Human eye detection method based on depth information and CNN | |
CN104376575A (en) | Pedestrian counting method and device based on monitoring of multiple cameras | |
CN107248174A (en) | A kind of method for tracking target based on TLD algorithms | |
GB2430736A (en) | Image processing | |
CN113255608B (en) | Multi-camera face recognition positioning method based on CNN classification | |
CN106971381A (en) | A kind of wide angle camera visual field line of demarcation generation method with the overlapping ken | |
CN115376109B (en) | Obstacle detection method, obstacle detection device, and storage medium | |
CN107492080A (en) | Exempt from calibration easily monocular lens image radial distortion antidote |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
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 |