CN103530879B - Pedestrian's color extraction method under special scenes - Google Patents
Pedestrian's color extraction method under special scenes Download PDFInfo
- Publication number
- CN103530879B CN103530879B CN201310481831.0A CN201310481831A CN103530879B CN 103530879 B CN103530879 B CN 103530879B CN 201310481831 A CN201310481831 A CN 201310481831A CN 103530879 B CN103530879 B CN 103530879B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- special scenes
- prospect
- image
- sample
- 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
Abstract
The invention provides a kind of pedestrian's color extraction method under special scenes, combine the pedestrian's Sample Storehouse collected under special scenes first by common row people's Sample Storehouse and use support vector machine to train the pedestrian dummy under special scenes；Then the video flowing of shooting under described special scenes uses mixed Gauss model carry out color modeling and prospect cutting, obtains distinguishing the prospect binary picture of background and prospect.Model and the gradient orientation histogram feature of the good pedestrian of training in advance that support vector machine combines is used in the region that prospect binary picture indicates in original image, different zoom is than lower retrieval pedestrian, merge the identical pedestrian target retrieved, it is combined result in conjunction with foreground segmentation result and accurately extracts the Zone Full of pedestrian, to this extracted region color histogram, obtain colouring information.The present invention can detect pedestrian in the video flowing of the automatically shooting of the photographic head from a certain scene, and extracts the colouring information of pedestrian.
Description
Technical field
The present invention relates to a kind of method of pedestrian's color extraction in video flowing, belong to computer vision, digital image processing field.
Background technology
In computer, field of processors, along with Moore's Law is constantly confirmed by scientific and technological progress, microcomputer capability improving is abnormal rapidly, and then increasing video monitoring system is applied in the daily management in city.On the one hand these systems produce monitor video contain substantial amounts of useful information, these information may be used for public security department solve a case, traffic department commander, planning department with reference to etc.；But on the other hand, the technological means extracting useful information from these massive video of current automatization is short of the most very much.
Summary of the invention
The present invention is based on computer vision and the latest developments of digital image processing field and video monitoring system problems faced instantly, the method proposing pedestrian's color extraction under a kind of special scenes, in the video flowing that the photographic head under a certain scene shoots, detect pedestrian, and extract the colouring information of pedestrian.
The technical scheme provided according to the present invention, the pedestrian's color extraction method under described special scenes combines the pedestrian's Sample Storehouse collected under special scenes and uses support vector machine to train the pedestrian dummy under special scenes first by common row people's Sample Storehouse；Then the video flowing of shooting under described special scenes use mixed Gauss model carry out color modeling and prospect cutting, obtain distinguishing the prospect binary picture of background and prospect, model and the gradient orientation histogram feature of the good pedestrian of training in advance that support vector machine combines is used in the region that prospect binary picture indicates in original image, different zoom is than lower retrieval pedestrian, merge the identical pedestrian target retrieved, it is combined result in conjunction with foreground segmentation result and accurately extracts the Zone Full of pedestrian, to this extracted region color histogram, obtain colouring information.
Further, described common row people's Sample Storehouse uses INRIA Sample Storehouse, comprises the positive sample of pedestrian's picture of 1805 64*128 resolution, and 1000 pictures not comprising pedestrian；The pedestrian's Sample Storehouse collected under described special scenes includes the positive sample that number is 1805 gathered under special scenes, and the picture not comprising pedestrian of 1000 special scenes；Amounting to 3610 positive samples, 2000 pictures not comprising pedestrian are as the material randomly selecting negative sample.
Further, the method for the pedestrian dummy under described training special scenes is:
Step 1, in described 2000 pictures not comprising pedestrian, win the sub-pictures of 3610 64*128 resolution at random as negative sample；
Step 2, calculate the gradient orientation histogram eigenvalue of positive and negative samples respectively, store in positive and negative samples eigenvalue file；The parameter of gradient direction eigenvalue calculation is: minimum lattice size is 6*6, block size is 3*3 minimum lattice, block overlapping degree is the block size of 1/2nd, and the post interval size of gradient orientation histogram is 20 degree, i.e. gradient is classified according to 18 regions on 0～360 degree of interval；
Step 3, respectively positive and negative samples is inputted be trained into SVM model plus positive and negative labelling, obtain representing the disaggregated model of pedestrian's feature；
In step 4, picture from remaining 1000 special scenes not comprising pedestrian, random selection goes out the negative sample of 1805 64*128, uses the disaggregated model produced in step 3 to carry out classification and obtains temporary pattern；
Step 5, the negative sample that the temporary pattern obtained in step 4 is categorized into positive result add to the negative sample produced in step 1；
Step 6, circulation perform step 2, obtain final disaggregated model 3,4,5 twice.
It is an advantage of the current invention that: the method can utilize the powerful calculating ability that computer is growing, the video flowing of the automatically shooting of the photographic head from a certain scene detects pedestrian, and extract the colouring information of pedestrian, and pedestrian's color above the waist can be drawn respectively, the concrete location information such as lower part of the body color.
Accompanying drawing explanation
Fig. 1 is algorithm overall flow figure.
Detailed description of the invention
Below in conjunction with drawings and Examples, the present invention is further described.
The present invention combines the pedestrian's Sample Storehouse collected under special scenes and uses SVM to train the pedestrian dummy under special scenes first by common row people's Sample Storehouse；Then the background to video flowing uses mixed Gauss model to carry out color modeling and prospect cutting, obtains distinguishing the prospect binary picture of background and prospect.Model and the gradient orientation histogram feature of the good pedestrian of training in advance that support vector machine combines is used in the region that prospect binary picture indicates in original image, different zoom is than lower retrieval pedestrian, merge the identical pedestrian target retrieved, it is combined result in conjunction with foreground segmentation result and accurately extracts the Zone Full of pedestrian, results area is extracted color histogram, obtains colouring information.
As it is shown in figure 1, whole algorithm is divided into two subprocess: model training, pedestrian are split and color extraction.Model training subprocess is that the sample gathered in advance according to system uses support vector machine to be trained, and obtains the pedestrian dummy under special scenes；First pedestrian detection and color extraction subprocess carry out background modeling according to video flowing, it is partitioned into prospect, pedestrian target is retrieved according to different zoom ratio the most in the foreground according to model, finally according to prospect with the pedestrian target Accurate Segmentation pedestrian retrieved at Zone Full in the picture, extract the colouring information i.e. colouring information of pedestrian target in this region.
1) model training.
A) training sample database prepares.
First selecting Sample Storehouse based on " INRIA " Sample Storehouse, this Sample Storehouse comprises the positive sample of pedestrian's picture of 1805 64*128 resolution from various scenes are extracted, and 1000 photos not comprising pedestrian；In addition to realize more preferable pedestrian detection effect, in addition it is also necessary to gather the positive sample that number is 1805 under specific occasion, 3610 positive samples altogether；Gathering the photo not comprising pedestrian of 1000 special scenes, 2000 images not comprising pedestrian are as the material randomly selecting negative sample altogether simultaneously.
B) model training.
1, do not comprise at 2000 INRIA in the picture of pedestrian and win the sub-pictures of 3610 64*128 resolution at random as negative sample.
2, calculate the gradient orientation histogram eigenvalue of positive negative sample respectively, store in positive and negative sample characteristics file.The parameter of gradient direction eigenvalue calculation is: minimum lattice size is 6*6, block size is 3*3 minimum lattice, block overlapping degree is the block size of 1/2nd, and the post interval size of gradient orientation histogram is 20 degree, i.e. gradient is classified according to 18 regions on 0～360 degree of interval.
3, positive negative sample adding, positive negative flag inputs respectively to be trained into SVM model, obtain representing the disaggregated model of pedestrian's feature.
4, in the picture from remaining 1000 special scenes not comprising pedestrian, random selection goes out the negative sample of 1805 64*128, uses the disaggregated model produced in step 3 to classify.
The negative sample that 5, model M1 in step 4 is categorized into positive result adds the negative sample produced to step 1.
6, circulation performs step 2, obtains final disaggregated model M 3,4,5 twice.
2) pedestrian's segmentation and color extraction.
A) foreground segmentation.
It is modeled the every two field picture in original video image analyzing, judgement prospect and the regional location of background, specifically include employing mixed Gauss model, in conjunction with color and the gradient of image, it is modeled the every two field picture in original video image analyzing, it is determined that prospect and the regional location of background.
Regional location according to described prospect with background carries out binary conversion treatment to described every two field picture, sets up the binary map for labelling prospect with position, background area, including:
Every two field picture for original video image sets up the pending image that a corresponding frame resolution is identical；
By in described pending image, the pixel corresponding to corresponding original video image foreground area gives white colour, and by described pending image, the pixel corresponding to corresponding original video image background area gives black colour, it is thus achieved that primary binary map；
Described primary binary map is carried out erosion operation, removes noise；
Primary binary map through removal noise processed is carried out dilation operation, it is thus achieved that the final binary map for labelling prospect Yu position, background area.
Calculate a rectangle, enable all foreground area to be included, record the position of rectangle frame.
B) pedestrian detection.
Through the exhaustive search method frequently with pyramid, Wo Menshe during detectionpsizeRepresent picture size,stepRepresent scaling step-length (each scaling),winsizeRepresent detection window size；Respectively by image to be detected with natural number (1,2,3 ...) multiplying power carries out reducing (until image higher primary school to be detected in 128 or wide is less than 64 stoppings), and on the image that each step reduces, carry out exhaustive search, will search window be that stepping is slided on detection image according to 1 pixel, and using disaggregated model to classify the image of each position, the preliminary classification obtained is the position candidate of pedestrian.
One pedestrian may be detected repeatedly on different yardsticks, and pedestrian duplicate detection gone out merges the pedestrian position finally detected.
C) accurately pedestrian is split and color feature extracted.
The pedestrian area that the foreground area obtained by foreground segmentation and pedestrian detection step detect projects in image to be detected, can obtain pedestrian's precise region in original region to be detected.
D) pedestrian's color and other feature extractions.
According in previous step to pedestrian's precise region in image to be detected, calculate the color histogram in this region, i.e. can obtain the colouring information of pedestrian.Can indicate on the precise region of pedestrian, it is also possible to extract other arbitrarily known to may be used for describing the characteristics of image of object.
Use the present invention accurate pedestrian segmentation result, can more other can show the eigenvalue of pedestrian's surface character at the place of extraction.In order to realize above-mentioned target, the foreground detection algorithm focusing on combining under complex scene up-to-date in computer vision of the present invention, pedestrian detection algorithm, composition is capable of pedestrian's color extraction method of above-mentioned target.The present invention has a characteristic that
1, the region that pedestrian is shared in the picture it is partitioned into accurately.
2, region shared by pedestrian's upper part of the body and the lower part of the body can be accurately positioned out.
3, can the accurately color component of pedestrian's whole body and the color component of special area at extraction.
4, combine original image and can extract the textural characteristics of place's pedestrian area.
5, extracting a kind of application that color is only this method, the present invention can also be used for extracting other features that may be used for describing pedestrian.
Claims (1)
1. the pedestrian's color extraction method under special scenes, is characterized in that: combines the pedestrian's Sample Storehouse collected under special scenes first by common row people's Sample Storehouse and uses support vector machine to train the pedestrian dummy under special scenes；Then use mixed Gauss model to carry out color modeling and foreground segmentation the video flowing of shooting under described special scenes, obtain distinguishing the prospect binary picture of background and prospect；Model and the gradient orientation histogram feature of the good pedestrian of training in advance that support vector machine combines is used in the region that prospect binary picture indicates in original image, different zoom is than lower retrieval pedestrian, merge identical retrieval result, merge the identical pedestrian target retrieved, it is combined result in conjunction with foreground segmentation result and accurately extracts the Zone Full of pedestrian, to this extracted region color histogram, obtain colouring information；
Wherein, described common row people's Sample Storehouse uses INRIA Sample Storehouse, comprises the positive sample of pedestrian's picture of 1805 64*128 resolution, and 1000 pictures not comprising pedestrian；The pedestrian's Sample Storehouse collected under described special scenes includes the positive sample that number is 1805 gathered under special scenes, and the picture not comprising pedestrian of 1000 special scenes；Amounting to 3610 positive samples, 2000 pictures not comprising pedestrian are as the material randomly selecting negative sample；
The method of the pedestrian dummy under described training special scenes is:
Step 1, in described 2000 pictures not comprising pedestrian, win the sub-pictures of 3610 64*128 resolution at random as negative sample；
Step 2, calculate the gradient orientation histogram eigenvalue of positive and negative samples respectively, store in positive and negative samples eigenvalue file；The parameter of gradient direction eigenvalue calculation is: minimum lattice size is 6*6, block size is 3*3 minimum lattice, block overlapping degree is the block size of 1/2nd, and the post interval size of gradient orientation histogram is 20 degree, i.e. gradient is classified according to 18 regions on 0～360 degree of interval；
Step 3, respectively positive and negative samples is inputted be trained into SVM model plus positive and negative labelling, obtain representing the disaggregated model of pedestrian's feature；
In step 4, picture from remaining 1000 special scenes not comprising pedestrian, random selection goes out the negative sample of 1805 64*128, uses the disaggregated model produced in step 3 to carry out classification and obtains temporary pattern；
Step 5, the negative sample that the temporary pattern obtained in step 4 is categorized into positive result add to the negative sample produced in step 1；
Step 6, circulation perform step 2, obtain final disaggregated model 3,4,5 twice；
The method of described foreground segmentation is:
It is modeled the every two field picture in original video image analyzing, judgement prospect and the regional location of background, specifically include employing mixed Gauss model, in conjunction with color and the gradient of image, it is modeled the every two field picture in original video image analyzing, it is determined that prospect and the regional location of background；
Regional location according to described prospect with background carries out binary conversion treatment to described every two field picture, sets up the binary map for labelling prospect with position, background area, including:
Every two field picture for original video image sets up the pending image that a corresponding frame resolution is identical；
By in described pending image, the pixel corresponding to corresponding original video image foreground area gives white colour, and by described pending image, the pixel corresponding to corresponding original video image background area gives black colour, it is thus achieved that primary binary map；
Described primary binary map is carried out erosion operation, removes noise；
Primary binary map through removal noise processed is carried out dilation operation, it is thus achieved that the final binary map for labelling prospect Yu position, background area；
Calculate a rectangle, enable all foreground area to be included, record the position of rectangle frame；
The method of described retrieval pedestrian is:
Respectively image to be detected is reduced with natural number multiplying power, until image higher primary school to be detected in 128 or wide is less than 64 stoppings, and on the image that each step reduces, carry out exhaustive search, will search window be that stepping is slided on detection image according to 1 pixel, and using disaggregated model to classify the image of each position, the preliminary classification obtained is the position candidate of pedestrian；
The method that described combination foreground segmentation result is combined the Zone Full that result accurately extracts pedestrian is:
The pedestrian area that the foreground area obtained by foreground segmentation and pedestrian detection step detect projects in image to be detected, obtains pedestrian's precise region in original region to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310481831.0A CN103530879B (en) | 2013-10-15 | 2013-10-15 | Pedestrian's color extraction method under special scenes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310481831.0A CN103530879B (en) | 2013-10-15 | 2013-10-15 | Pedestrian's color extraction method under special scenes |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103530879A CN103530879A (en) | 2014-01-22 |
| CN103530879B true CN103530879B (en) | 2016-08-17 |
Family
ID=49932858
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310481831.0A Active CN103530879B (en) | 2013-10-15 | 2013-10-15 | Pedestrian's color extraction method under special scenes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103530879B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107085713A (en) * | 2017-05-05 | 2017-08-22 | 中山大学 | End-to-end quick pedestrian recognition methods again based on correlation filtering |
| CN111461036B (en) * | 2020-04-07 | 2022-07-05 | 武汉大学 | Real-time pedestrian detection method using background modeling to enhance data |
| CN112203023B (en) * | 2020-09-18 | 2023-09-12 | 西安拙河安见信息科技有限公司 | Billion pixel video generation method and device, equipment and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102147869A (en) * | 2011-03-31 | 2011-08-10 | 上海交通大学 | Pedestrian detection method based on foreground analysis and pattern recognition |
| CN103034852A (en) * | 2012-12-28 | 2013-04-10 | 上海交通大学 | Specific color pedestrian detecting method in static video camera scene |
-
2013
- 2013-10-15 CN CN201310481831.0A patent/CN103530879B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102147869A (en) * | 2011-03-31 | 2011-08-10 | 上海交通大学 | Pedestrian detection method based on foreground analysis and pattern recognition |
| CN103034852A (en) * | 2012-12-28 | 2013-04-10 | 上海交通大学 | Specific color pedestrian detecting method in static video camera scene |
Non-Patent Citations (1)
| Title |
|---|
| 基于多特征的行人检测方法研究;陈刚;《中国优秀硕士学位论文全文数据库信息科技辑》;20130215;摘要，第4.4.1-4.4.2节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103530879A (en) | 2014-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Rapid localization and extraction of street light poles in mobile LiDAR point clouds: A supervoxel-based approach | |
| CN103530638B (en) | Method for pedestrian matching under multi-cam | |
| Wang et al. | Instance shadow detection | |
| CN102906790B (en) | Image information processing device | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| Ghaffarian | Automatic building detection based on supervised classification using high resolution Google Earth images | |
| CN108334847A (en) | A kind of face identification method based on deep learning under real scene | |
| CN102043953A (en) | Real-time-robust pedestrian detection method aiming at specific scene | |
| CN105160297A (en) | Masked man event automatic detection method based on skin color characteristics | |
| CN108537782A (en) | A method of building images match based on contours extract with merge | |
| CN107679470A (en) | A kind of traffic mark board detection and recognition methods based on HDR technologies | |
| CN102034267A (en) | Three-dimensional reconstruction method of target based on attention | |
| CN105184308B (en) | Remote sensing image building detection classification method based on global optimization decision | |
| CN105069816B (en) | A kind of method and system of inlet and outlet people flow rate statistical | |
| CN106228157A (en) | Coloured image word paragraph segmentation based on image recognition technology and recognition methods | |
| CN103971524A (en) | Traffic flow detection method based on machine vision | |
| CN103530879B (en) | Pedestrian's color extraction method under special scenes | |
| CN114565675A (en) | Method for removing dynamic feature points at front end of visual SLAM | |
| CN108898098A (en) | Early stage video smoke detection method based on monitor supervision platform | |
| Zhu et al. | Anomaly detection via 3D-HOF and fast double sparse representation | |
| Hara et al. | Exploring early solutions for automatically identifying inaccessible sidewalks in the physical world using google street view | |
| Patel et al. | A novel approach for detecting number plate based on overlapping window and region clustering for Indian conditions | |
| Wolf et al. | Applicability of neural networks for image classification on object detection in mobile mapping 3d point clouds | |
| Sun et al. | Unsupervised object extraction by contour delineation and texture discrimination based on oriented edge features | |
| Sun et al. | Multi-purpose SLAM framework for dynamic environment |
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 |