CN103049749B - The recognition methods again of human body under grid blocks - Google Patents
The recognition methods again of human body under grid blocks Download PDFInfo
- Publication number
- CN103049749B CN103049749B CN201210592918.0A CN201210592918A CN103049749B CN 103049749 B CN103049749 B CN 103049749B CN 201210592918 A CN201210592918 A CN 201210592918A CN 103049749 B CN103049749 B CN 103049749B
- Authority
- CN
- China
- Prior art keywords
- human body
- image
- region
- body image
- vector
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a kind of recognition methods again of human body under grid blocks, including: detect the human body image in video image；Described human body image is divided into multiple region；By in the multiple regions after segmentation, remove the region at grid barrier place；Determine the characteristic vector in each described region, multiple characteristic vectors are mated with the multiple reference vector in the data base gathered in advance；Using the human body image that in described data base, the match is successful as recognition result.By above-mentioned step, detection process eliminates the occlusion area that barrier is formed, human body image can be determined accurately in data base, it is to be determined to the label of the human body image gone out or ID are as the human body image detected.Thus the range of activity of everyone volume image can be grasped in video.
Description
Technical field
The present invention relates to field of video monitoring, in particular to a kind of recognition methods again of human body under grid blocks.
Background technology
At present to the human body recognition technology in video image, in identification process, owing to the environment of surrounding causes blocking human body, cause affecting recognition result, such as, there is the shelters such as fence in the security protection region of shooting, forms blocking such as lattice-shaped on the human body image of shooting.Existing video identification technology, the human body image in video can only be identified, the individuality of human body image can not be confirmed, under above-mentioned environment, the image of shooting can cause again the grid shelter to human body image, cause different recognition results, thus causing the motion track not distinguishing everyone volume image, it is impossible to determine the identity of human body image in current video.
Summary of the invention
It is desirable to provide a kind of recognition methods again of human body under grid blocks, with the problem that the individuality of human body image must not be confirmed by solution.
In an embodiment of the present invention, it is provided that a kind of recognition methods again of human body under grid blocks, including: detect the human body image in video image；Described human body image is divided into multiple region；By in the multiple regions after segmentation, remove the region at grid barrier place；Determine the characteristic vector in each described region, multiple characteristic vectors are mated with the multiple reference vector in the data base gathered in advance；Using the human body image that in described data base, the match is successful as recognition result.
By above-mentioned step, human body image can be determined in data base, it is to be determined to the human body image identity gone out is as the identity of the human body image detected.Thus the range of activity of people corresponding to everyone volume image can be grasped in video.Owing to detection process eliminating the occlusion area that barrier is formed, add the accuracy rate of identification.
Accompanying drawing explanation
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the application, and the schematic description and description of the present invention is used for explaining the present invention, is not intended that inappropriate limitation of the present invention.In the accompanying drawings:
Fig. 1 illustrates the flow chart of embodiment；
Detailed description of the invention
Below with reference to the accompanying drawings and in conjunction with the embodiments, the present invention is described in detail.Referring to Fig. 1, the step of embodiment includes:
S11: detect the human body image in video image；
S12: described human body image is divided into multiple region；
S13: by the multiple regions after segmentation, remove the region at grid barrier place；
Multiple characteristic vectors are mated by S14: determine the characteristic vector in each described region with the multiple reference vector in the data base gathered in advance；
S15: using the human body image that in described data base, the match is successful as recognition result.
By above-mentioned step, human body image can be determined in data base, it is to be determined to the human body image identity gone out is as the identity of the human body image detected.Thus people corresponding to everyone volume image and range of activity can be grasped in video.Owing to detection process eliminating the occlusion area that barrier is formed, add the accuracy rate of identification.
Preferably, in embodiment, the step of detection human body image includes: use Gaussian Background modeling to detect moving region in video.In order to eliminate noise, use corrosion and expansion algorithm that the foreground picture detected is filtered.Draw a circle to approve out by foreground picture region, as the scope of human detection.
In the moving region detected, use the object detecting method based on histograms of oriented gradients (HOG) and the support vector machine (latentSVM) with implicit parameter, the human body image in video is detected by different scale.
Preferably, in embodiment, when image is split, can adopt watershed algorithm that image is split.In the picture, choose point that gray value the is local minimum seed as watershed algorithm, the half-tone information of image is used watershed algorithm, is different regions by picture segmentation.
The formula of the gray scale calculating pixel is as follows: Y=0.2999R+0.5870G+0.1140B
Watershed algorithm segmentation image: watershed algorithm is the half-tone information according to image, and image carries out a kind of method of region segmentation.First all pixels in image are sorted from small to large according to gray value, using point that gray value is local minimum as seed points.In structure region, each seed points position.Process each pixel one by one according to gray value order from small to large afterwards, processed pixel is added among the region adjacent with it.After all pixels are all added into region, just obtain the segmentation information of image.The region of segmentation is generally the upper part of the body image of human body image, lower part of the body image and head, even can also have foot etc..
Adopt watershed algorithm specific implementation as follows:
M1M2 ... .MR represents image g(x, the set of the coordinate of local minizing point y).R is positive integer.
C (Mi) represents and the set of point in the local minimum Mi catchment basin being associated.
T [n]=(s, t) | (s, t) < n} represents and is positioned at plane g(x g, the set of the point below y)=n.S, t are coordinate points.
Cn (Mi)=C (Mi) ∩ T [n] represents the set by water logging no part of the n-th order section catchment basin.Mi=M1 ~ MR
Q represents the set of continuous component in T [n].Each continuous component q ∈ there are three kinds of possibilities
A () q ∩ C [n-1] is empty
B () q ∩ C [n-1] comprises a connected component in C [n-1].
C () q ∩ C [n-1] comprises C [n-1] more than one connected component.
When running into new minima, eligible (a), q is incorporated to c [n-1], constitutes c [n];
When Q is positioned at the catchment basin that some local minimum is constituted, eligible (b), q is incorporated to c [n-1] and constitutes c [n], when running into all or part of catchment basin of separation, eligible (c), set up dam at q.Dam is the demarcation line, edge of the image of two different colours.
End condition is n=max+1.The color interval of max pixel, for instance: in gray scale, 255 is the highest.
Preferably, the image after segmentation is eliminated over-segmentation: after obtaining image segmentation information, calculate the average gray in each region, compared by the average gray in adjacent region, when difference is less than threshold value 5, be one by two region merging technique.
Preferably, in embodiment, it is determined that the process of characteristic vector includes:
The image detected is converted to HSV form, and extracts distribution of color rectangular histogram.
From RGB color to the conversion in hsv color space, computing formula is as follows:
V=max
Wherein max=max (r, g, b), min=min (r, g, b).Such as, for the pixel that RGB color value is (0.1,0.2,0.5), the value in hsv color space is (225,0.8,0.5).
Calculate color histogram: for each pixel in image, its color is added up.Such as, v component is black less than threshold value 1, and v component is white more than threshold value 2 and s component less than threshold value 3, v component between threshold value 1 and threshold value 2 and v component be Lycoperdon polymorphum Vitt less than threshold value 3, other colors are colour.
For colour, being evenly dividing from 0 to 360 according to h component is 6 kinds of colors, namely [0,60), [60,120), [120,180), [180,240), [240,300), [300,360).
The color of each pixel being added up, and calculates each color proportion in each region of human body image, store successively in array x, the characteristic vector as image uses.
Such as, an image-region there are 10 pixels.Wherein black color dots and white point are respectively arranged with 3, other 4 points belong to color [60,120), then this region characteristic of correspondence vector is (0.3,0.3,0,0,0.4,0,0,0,0).
Preferably, in embodiment, the reference vector in described data base is determined by following steps:
Gather several video images of everyone volume image in advance；
By several video images described, it is determined that go out multiple regions of this human body image and a stack features vector corresponding with each region, as the reference vector that this region is corresponding.
For the blocking human body image that grid is formed, the feature of grid can be set in advance in a program, for the barrier of grid shape, it is characterized by porous.Therefore, the region that numbers of hole among the region split only need to exceed threshold value excludes, and remaining region is human region.Such as: outdoor grid fence, the fence etc. in roadside.
The method that this process can be passed through to cluster realizes, for instance: use K-means (K average) scheduling algorithm.
When using K-means training, everyone color histogram in volume image region obtained is clustered as characteristic vector, obtain the cluster centre of characteristic vector and the zone sample that each cluster centre comprises in detection process.
K mean algorithm needs one parameter k of input and several characteristic vectors.Calculated by K mean algorithm and these characteristic vectors can be divided into k class and the sample that each apoplexy due to endogenous wind comprises.In this manner it is possible to the sample of input is divided into k class, each class represent a human body image.
The regional characteristic of correspondence vector that cluster centre obtains each class stores in data base.
Above-mentioned matching process includes:
Each characteristic vector corresponding to each described region of computing respectively with the distance of the reference vector of the regional of everyone volume image in described data base；
Multiple distance-taxis that each characteristic vector is obtained, it is determined that go out two minimum distance d1 and d2；Wherein, d1 < d2；
If described 1.5d1 < d2, it is determined that this characteristic vector and the reference vector of d1 described in computing match.
Determine the human body image that the region of the reference vector closest with each described characteristic vector is corresponding, and the summation of number of times that the reference vector adding up the regional of everyone volume image corresponding is matched；
Find out label or the ID of the human body image of the unique and the highest value of the number of times summation being determined, as the described human body image that the match is successful.
Wherein, the reference vector calculating minimum Euclidean distance it is used for as apart from the highest reference vector.The formula of Euclidean distance is as follows:
Wherein d is the distance of characteristic vector and reference vector, and x is the characteristic vector of image, and X is the reference vector that training obtains, and i represents the figure place of characteristic vector or reference vector, and N is the dimension of characteristic vector or reference vector.
Assuming that human body image is divided into some regions, wherein ith zone is identified as block pi, come from s in data baseiIndividual human body image.To siCarry out statistics with histogram, and whole human body image is classified as the model corresponding to component maximum in rectangular histogram.
Such as: if one has 5 human body image samples, multiple regions that each sample comprises by a human body image multiple reference vector corresponding to difference.
The human body image detected is divided into 3 regions, totally 3 characteristic vectors；Data base includes 5 human body image samples, and each sample includes 3 regions, then have 15 regions, the corresponding reference vector in each region.Calculate the distance of each characteristic vector and 15 reference vector detected, obtain 5 groups of data.
Often group data include 15 distances, find minimum two distance, d1 and d2, and meet 1.5d1 < d2, then it is assumed that match reference vector.
Add up the number of times that each reference vector of everyone volume image is matched.Such as: the characteristic vector detecting certain region is (1,0,0,0,0,0,0,0,0), two reference vector respectively (0.8,0,0,0,0,0,0,0,0.2) closest with it and (0.5,0.5,0,0,0,0,0,0,0).Then can calculate and obtain d1≈ 0.283, d2≈ 0.707, and 1.5d1<d2.Determine that this characteristic vector and the reference vector of d1 described in computing match.The reference vector of d1 described in computing is the human body image of sample 1, then the human body image of sample 1 is for identifying successful human body image.
The region that regional is respectively identified as in each sample following；As: sample 1, sample 1, sample 2, then statistic histogram is (2,1), and sample 1 is the highest and unique sample, the human body image that the human body image being detected finally is identified as corresponding to sample 1 again.
It addition, in order to realize accurate coupling, the image zooming-out ORB characteristic point to the human body image identified and sample, use hamming distance that characteristic point is mated, and use RANSAC algorithm to eliminate erroneous matching.Determine whether that the match is successful according to final matching result.
Obviously, those skilled in the art should be understood that, each module of the above-mentioned present invention or each step can realize with general calculation element, they can concentrate on single calculation element, or it is distributed on the network that multiple calculation element forms, alternatively, they can realize with the executable program code of calculation element, thus, can be stored in storage device is performed by calculation element, or they are fabricated to respectively each integrated circuit modules, or the multiple modules in them or step are fabricated to single integrated circuit module realize.So, the present invention is not restricted to the combination of any specific hardware and software.
The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, for a person skilled in the art, the present invention can have various modifications and variations.All within the spirit and principles in the present invention, any amendment of making, equivalent replacement, improvement etc., should be included within protection scope of the present invention.
Claims (9)
1. the recognition methods again of the human body under grid blocks, it is characterised in that including:
Detect the human body image in video image；
The described human body image detected in video image, including:
In the moving region detected, use the object detecting method based on histograms of oriented gradients and the support vector machine with implicit parameter, different scale detects the human body image in video；
Described human body image is divided into multiple region；
By in the multiple regions after segmentation, remove the region at grid barrier place；
Determine the characteristic vector in each described region, multiple characteristic vectors are mated with the multiple reference vector in the data base gathered in advance；
Using the human body image that in described data base, the match is successful as recognition result；
Image zooming-out ORB characteristic point to the human body image identified and sample, uses hamming distance that characteristic point is mated, and uses RANSAC algorithm to eliminate erroneous matching, determines whether that the match is successful according to final matching result.
2. method according to claim 1, it is characterised in that described cutting procedure includes:
Local minimum in human body image is selected as seed, to adopt watershed algorithm to be divided into multiple region.
3. method according to claim 2, it is characterised in that also include:
Relatively the color gray scale of adjacent area, when difference is less than threshold value, merges described adjacent area.
4. method according to claim 2, it is characterised in that the described process determining characteristic vector includes:
The image in described region is converted to HSV form；
Add up the pixel quantity of shades of colour in the region of described HSV form；
Pixel quantity according to described shades of colour determines a stack features vector corresponding with this region.
5. method according to claim 4, it is characterised in that the reference vector in described data base is determined by following steps:
Gather several video images of everyone volume image in advance；
By several video images described, it is determined that go out multiple regions of everyone volume image and a stack features vector corresponding with each region, as the reference vector that this region is corresponding；
Described matching process includes:
Each characteristic vector corresponding to each described region of computing respectively with the distance of the reference vector of the regional of everyone volume image in described data base；
Multiple distance-taxis that each characteristic vector is obtained, it is determined that go out two minimum distance d1 and d2；Wherein, d1 < d2；
If described 1.5d1 < d2, it is determined that this characteristic vector and the reference vector of d1 described in computing match.
6. method according to claim 5, it is characterised in that
Determine the human body image that the region of the reference vector closest with each described characteristic vector is corresponding, and the summation of number of times that the reference vector adding up the regional of everyone volume image corresponding is matched；
Find out label or the ID of the human body image of the unique and the highest value of the number of times summation being determined, as the described human body image that the match is successful.
7. method according to claim 5, it is characterised in that also include: adopt distance described in following Euclidean distance formula operation；
Wherein d is the distance of characteristic vector and reference vector, and x is the characteristic vector of image, and X is the reference vector that training obtains, and i represents the figure place of characteristic vector or reference vector, and N is the dimension of characteristic vector or reference vector.
8. method according to claim 6, it is characterised in that also include:
Without the match is successful, then the characteristic vector of the regional of the described human body image detected is joined described data base as new reference vector.
9. method according to claim 1, it is characterised in that also include:
Current frame image and before video image in, adopt and minimum state, with color receptacle frame residence, this human body image detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210592918.0A CN103049749B (en) | 2012-12-30 | 2012-12-30 | The recognition methods again of human body under grid blocks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210592918.0A CN103049749B (en) | 2012-12-30 | 2012-12-30 | The recognition methods again of human body under grid blocks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103049749A CN103049749A (en) | 2013-04-17 |
| CN103049749B true CN103049749B (en) | 2016-06-29 |
Family
ID=48062381
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210592918.0A Expired - Fee Related CN103049749B (en) | 2012-12-30 | 2012-12-30 | The recognition methods again of human body under grid blocks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103049749B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103729462B (en) * | 2014-01-13 | 2016-09-14 | 武汉大学 | A kind of pedestrian retrieval method blocked based on rarefaction representation process |
| CN109478328A (en) * | 2016-09-30 | 2019-03-15 | 富士通株式会社 | Method for tracking target, device and image processing equipment |
| CN108154171B (en) * | 2017-12-20 | 2021-04-23 | 北京奇艺世纪科技有限公司 | Figure identification method and device and electronic equipment |
| CN111832361B (en) * | 2019-04-19 | 2023-08-29 | 杭州海康威视数字技术股份有限公司 | Pedestrian re-identification method and device and computer equipment |
| CN111077990B (en) * | 2019-06-03 | 2024-03-19 | 广东小天才科技有限公司 | Method for determining content to be read on spot and learning equipment |
| CN111339996B (en) * | 2020-03-20 | 2023-05-09 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for detecting static obstacle |
| CN114549921B (en) * | 2021-12-30 | 2023-04-07 | 浙江大华技术股份有限公司 | Object recognition method, electronic device, and computer-readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101311947A (en) * | 2008-06-12 | 2008-11-26 | 浙江大学 | Real time intelligent control method based on natural video frequency |
| CN101325690A (en) * | 2007-06-12 | 2008-12-17 | 上海正电科技发展有限公司 | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow |
| CN101710422A (en) * | 2009-12-11 | 2010-05-19 | 西安电子科技大学 | Image segmentation method based on overall manifold prototype clustering algorithm and watershed algorithm |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100734849B1 (en) * | 2005-11-26 | 2007-07-03 | 한국전자통신연구원 | Method for recognizing face and apparatus thereof |
| KR100904916B1 (en) * | 2008-06-19 | 2009-07-01 | 주식회사 다우엑실리콘 | System and method for recognition of face |
| CN101329725B (en) * | 2008-07-30 | 2010-10-06 | 电子科技大学 | Method for dividing fingerprint image based on gradient projection and morphology |
| CN101577052B (en) * | 2009-05-14 | 2011-06-08 | 中国科学技术大学 | Device and method for detecting vehicles by overlooking |
-
2012
- 2012-12-30 CN CN201210592918.0A patent/CN103049749B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101325690A (en) * | 2007-06-12 | 2008-12-17 | 上海正电科技发展有限公司 | Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow |
| CN101311947A (en) * | 2008-06-12 | 2008-11-26 | 浙江大学 | Real time intelligent control method based on natural video frequency |
| CN101710422A (en) * | 2009-12-11 | 2010-05-19 | 西安电子科技大学 | Image segmentation method based on overall manifold prototype clustering algorithm and watershed algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 基于SIFT的人脸识别研究;罗佳;《中国优秀硕士学位论文全文数据库信息科技辑（月刊）》;20110515(第05期);24-40 * |
| 基于改进分水岭的棉花图像分割方法;任磊等;《计算机工程与应用》;20121201;第48卷(第34期);207-211 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103049749A (en) | 2013-04-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103049749B (en) | The recognition methods again of human body under grid blocks | |
| CN103093274B (en) | Method based on the people counting of video | |
| CN103065126B (en) | Re-identification method of different scenes on human body images | |
| CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
| WO2017190574A1 (en) | Fast pedestrian detection method based on aggregation channel features | |
| Ryan et al. | Scene invariant multi camera crowd counting | |
| CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
| CN107392968B (en) | The image significance detection method of Fusion of Color comparison diagram and Color-spatial distribution figure | |
| CN107886507B (en) | A kind of salient region detecting method based on image background and spatial position | |
| CN101276461A (en) | Method for increasing video text with edge characteristic | |
| CN106778635A (en) | A kind of human region detection method of view-based access control model conspicuousness | |
| CN102915433A (en) | Character combination-based license plate positioning and identifying method | |
| CN105069816B (en) | A kind of method and system of inlet and outlet people flow rate statistical | |
| CN104966305A (en) | Foreground detection method based on motion vector division | |
| CN104217213A (en) | Medical image multi-stage classification method based on symmetry theory | |
| Patel et al. | Automatic licenses plate recognition | |
| CN104217206A (en) | Real-time attendance counting method based on high-definition videos | |
| CN103390151A (en) | Face detection method and device | |
| Mammeri et al. | North-American speed limit sign detection and recognition for smart cars | |
| Bullkich et al. | Moving shadow detection by nonlinear tone-mapping | |
| CN105354570A (en) | Method and system for precisely locating left and right boundaries of license plate | |
| CN103065129B (en) | Giant panda is known method for distinguishing | |
| CN106056078A (en) | Crowd density estimation method based on multi-feature regression ensemble learning | |
| CN103077376B (en) | Method for distinguishing is known again based on the human body image in video image | |
| US20160283805A1 (en) | Method and device for classifying an object in an image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160629 Termination date: 20191230 |