CN109003289B - Target tracking rapid initialization method based on color label - Google Patents
Target tracking rapid initialization method based on color label Download PDFInfo
- Publication number
- CN109003289B CN109003289B CN201711334249.6A CN201711334249A CN109003289B CN 109003289 B CN109003289 B CN 109003289B CN 201711334249 A CN201711334249 A CN 201711334249A CN 109003289 B CN109003289 B CN 109003289B
- Authority
- CN
- China
- Prior art keywords
- color
- image
- background
- target
- label
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- 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
- G06T2207/10024—Color image
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A target tracking rapid initialization method based on color labels. 1. The invention divides the color into N grades in the Lab color picture; 2. the target rectangular box is extracted and the color space is converted to the Lab color space. 3. Dividing the image into subareas; 4. calculating the color grade of the sub-region color block; 5. finding all connected domains connected with the background elements; 6. eliminating the connected domain; 7. obtaining an image with prominent foreground information; 8. performing AND operation on the obtained result image with the outstanding foreground information and the original image picture; 9. and outputting the image without the background image as a result, and performing target tracking initialization and feature extraction. Compared with other methods, the method can enable the extracted target characteristics to be always stronger than the background characteristics, has the advantages of high speed, simple implementation, good effect and good flexibility, does not depend on strong hardware support, and is low in cost.
Description
Technical Field
The invention relates to the technical field of image processing and analysis, in particular to a method for processing target tracking by using a color label method, which can quickly extract image foreground information, destroy background information and improve the tracking performance of a target under a complex background.
Background
In recent years, the target tracking technology is receiving more and more attention along with the development of urban security, and the target tracking technology is widely applied to places such as large squares, school parks or frontier sea areas. The mainstream methods of the target tracking technology include a generating formula and a discriminating formula; the generating method is mainly used for modeling a target area of a current frame, and the area which is most similar to the model and searched by the next frame is the predicted position; the discriminant method mainly uses a target area of a current frame as a positive sample and a background area as a negative sample, and searches an optimal area for the next frame by using a trained classifier through a machine learning training classifier. Although the research of the target tracking technology is more and more mature, the interference of factors such as variable environment, complex background and the like in an actual monitoring scene has a great influence on the performance of the target tracking method.
In order to solve the problem, people focus on two aspects of selection of features and classifiers. On one hand, people try to combine multiple features, design features with higher dimensionality to express target information more accurately, and improve tracking performance. On the other hand, starting from the classifier, designing a classifier with better performance is also a way to improve the tracking accuracy. Moreover, the deep learning is introduced into the target tracking field, and the target tracking performance is greatly improved by starting from two aspects of characteristics and a classifier. Although the method can improve certain performance, the calculation overhead is huge, and the real-time requirement of video monitoring is difficult to meet.
Disclosure of Invention
For target tracking, good features often have a direct impact on the tracking result. In the initialization process of target tracking, a target rectangular frame needs to be specified in a target detection or manual extraction mode, certain features are extracted from an image in the rectangular frame, and then a tracking task is completed through processing, identification and matching of the features. The rectangular frame contains rich background information in addition to the information of the target. Under the condition that the background is simpler, the feature expression of the target in the rectangular frame is stronger than that of the background, the extracted feature discrimination is stronger, and the tracking effect is good. However, the background in the actual scene is often complex, the feature expression of the target in the rectangular frame may be weaker than the background feature, and the degree of distinguishing the extracted features is weak, resulting in a tracking failure. Therefore, even if the extracted feature has a higher dimension and is more capable of expressing the object information, the obtained feature is erroneous if the object in the rectangular frame cannot be distinguished from the background.
In order to solve the problem, the invention starts from the target tracking initial image, quickly extracts the target information in the rectangular frame by the color label technology, destroys the background information in the rectangular frame, highlights the target characteristics, weakens the background characteristics, enables the expression of the target characteristics to be always stronger than that of the background characteristics, and improves the tracking performance under the complex background. The method has the advantages of high speed, simple realization and flexible use, and can improve the tracking performance of various target tracking methods.
The technical scheme of the invention is as follows:
the color labels, as the name implies, fixedly divide all colors in the color space into several color categories, and assign a color label to each color category.
The first step is as follows: the invention divides the color into N grades in the Lab color picture, each color grade is endowed with a label numerical value, the color label is 1-N in total, namely divided into N colors, namely the N colors approximately represent all the colors in one picture;
the second step is as follows: dividing the image into subareas, longitudinally and transversely cutting one image into a plurality of small subareas, wherein each subarea is M pixel by M pixel, and the pixel color of each small subarea is kept uniform as much as possible; establishing an initial color label graph;
the third step: extracting a rectangular frame containing a target by a target detection method, performing color space conversion in the rectangular frame, and converting a color space formed by a RGB color system of a three-color channel into a Lab pixel space;
the fourth step: calculating the color grade of each small sub-region color block, respectively counting the average value of pixels in each sub-region, performing Euclidean distance calculation with the N calibrated color labels, and selecting the color label with the minimum Euclidean distance as the color label corresponding to the current sub-region;
the fifth step: setting: the outermost element of the color label graph is a background; starting from each background element on the outermost periphery, searching from the boundary to the inside by searching for the connectivity between the elements, and expanding the subareas one by one until all connected domains connected with the background elements are searched;
a sixth step: eliminating all searched connected domains to obtain a color label map only containing foreground information;
a seventh step of: eliminating all background information to obtain an image with prominent foreground information;
an eighth step: performing AND operation on the image with the outstanding foreground information and the original image picture to obtain the original image which only retains the foreground information after the background image is removed as a result;
a ninth step: and outputting the original image only retaining the foreground information after removing the background information as a result, and performing target tracking initialization and feature extraction. The extracted features only contain target foreground information, so that background interference is reduced, and the target tracking performance under a complex background can be quickly and effectively improved.
The principle of the invention is as follows:
1. dividing colors in the color space into n color labels, and judging the color label to which any color in the color space belongs by calculating the Euclidean distance. And simultaneously, cutting the image into sub-regions with the size of m × m, counting the color mean value of each sub-region, and calculating the color label to which the sub-region belongs to form a color label graph.
2. And searching a connected domain connected with the background element by taking the outermost element of the color label graph as a background in an outside-in searching expansion mode, and removing the connected domain to obtain an image only containing target information.
According to the method, the color label expansion processing is carried out on the target tracking initialization image, the target information is extracted, the background information is damaged, the characteristic expression of the target is always stronger than that of the background, and the tracking performance under the complex background is improved. Compared with other methods, the method destroys background information, only reserves target information, enables target characteristics extracted by tracking initialization to be stronger than background characteristics, and simultaneously reduces extraction time of the characteristics. Therefore, the method has the advantages of high speed, simple realization, good effect, good flexibility, no dependence on strong hardware support and low cost.
Drawings
FIG. 1 is a color label classification chart according to the present invention:
FIG. 2 is a schematic diagram of a target-marking rectangular frame in a monitoring screen according to the present invention;
FIG. 3 is a schematic view of a sub-region cutting method according to the present invention;
FIG. 4 is a schematic diagram of a color label corresponding to a current sub-region of a picture according to the present invention;
FIG. 5 is a schematic diagram of a frame for performing background calibration on a picture according to the present invention;
FIG. 6 is a schematic diagram illustrating the expansion of the connectivity of the background area elements of the frame according to the present invention;
FIG. 7 is a diagram illustrating foreground information after background region elements are eliminated according to the present invention;
FIG. 8 is a diagram illustrating the result of the background destruction with foreground highlighting;
FIG. 9 is a diagram illustrating the results of the present invention and the artwork and after operation.
Detailed Description
Example 1: (vehicle)
The first step is as follows: because the area occupied by the vehicle in the monitoring picture is large, the number of pixels is large, and the foreground and the background are easy to distinguish, the scale of the color label can be properly reduced, and the pixels of each sub-area can be increased. The invention divides the colors into 10 grades in the Lab color picture, each color grade is assigned with a label numerical value, and the color labels are 10 in total, namely, divided into 10 colors, namely, the 10 colors approximately represent all the colors in one picture.
The second step is as follows: dividing the image into subregions, carrying out subregion longitudinal and transverse cutting on one picture, dividing the picture into a plurality of small subregions, wherein the size of each subregion is 8 x 8 pixels, and keeping the pixel color of each small subregion uniform; establishing an initial color label graph;
the third step: extracting a rectangular frame containing a target by a target detection method, performing color space conversion in the rectangular frame, and converting a color space formed by a RGB color system of a three-color channel into a Lab pixel space;
the fourth step: calculating the color grade of each small sub-region color block, respectively counting the average value of pixels in each sub-region, performing Euclidean distance calculation with the 10 calibrated color labels, and selecting the color label with the minimum Euclidean distance as the color label corresponding to the current sub-region;
the fifth step: setting: the outermost element of the color label graph is a background; starting from each background element on the outermost periphery, searching from the boundary to the inside by searching for the connectivity between the elements, and expanding the subareas one by one until all connected domains connected with the background elements are searched;
a sixth step: eliminating all searched connected domains to obtain a color label map only containing foreground information;
a seventh step of: eliminating all background information to obtain an image with prominent foreground information;
an eighth step: performing AND operation on the image with the outstanding foreground information and the original image picture to obtain an image with the background image removed as a result;
a ninth step: and outputting the image without the background image as a result, and performing target tracking initialization and feature extraction. The extracted features only contain target information, so that background interference is reduced, and the target tracking performance under a complex background can be quickly and effectively improved.
Example 2: (pedestrian)
The first step is as follows: because the area occupied by the pedestrian in the monitoring picture is small, the contained pixels are few, and the foreground and the background are difficult to distinguish, the scale of the color label can be enlarged properly, and the pixels of each sub-area can be reduced. The invention divides the colors into 15 grades in the Lab color picture, each color grade is assigned with a label numerical value, and the color labels are divided into 15 colors, namely 15 colors approximately represent all the colors in one picture.
The second step is as follows: dividing the image into subareas, vertically and horizontally cutting one picture into a plurality of small subareas, wherein the size of each subarea is 4 x 4 pixels, and the pixel color of each small subarea is kept uniform as much as possible; establishing an initial color label graph;
the third step: extracting a rectangular frame containing a target by a target detection method, performing color space conversion in the rectangular frame, and converting a color space formed by a RGB color system of a three-color channel into a Lab pixel space;
the fourth step: calculating the color grade of each small sub-region color block, respectively counting the average value of pixels in each sub-region, performing Euclidean distance calculation with the 15 calibrated color labels, and selecting the color label with the minimum Euclidean distance as the color label corresponding to the current sub-region;
the fifth step: setting: the outermost element of the color label graph is a background; starting from each background element on the outermost periphery, searching from the boundary to the inside by searching for the connectivity between the elements, and expanding the subareas one by one until all connected domains connected with the background elements are searched;
a sixth step: eliminating all searched connected domains to obtain a color label map only containing foreground information;
a seventh step of: eliminating all background information to obtain an image with prominent foreground information;
an eighth step: performing AND operation on the image with the outstanding foreground information and the original image picture to obtain an image with the background image removed as a result;
a ninth step: and outputting the image without the background image as a result, and performing target tracking initialization and feature extraction. The extracted features only contain target information, so that background interference is reduced, and the target tracking performance under a complex background can be quickly and effectively improved.
Example 3: (multiple persons)
The first step is as follows: because many people often occupy a relatively large area in the monitoring picture and contain a large number of pixels, but the area contains a plurality of pedestrians, and the pedestrians need to be refined and distinguished at the same time, the scale of the color label can be properly enlarged, and the pixels of each sub-region can be further reduced. The invention divides the colors into 25 grades in a Lab color picture, each color grade is assigned with a label numerical value, and the color labels are 25, namely divided into 25 colors, namely, the 25 colors approximately represent all the colors in one picture.
The second step is as follows: dividing the image into subregions, carrying out subregion longitudinal and transverse cutting on one picture, dividing the picture into a plurality of small subregions, wherein the size of each subregion is 2 x 2 pixels, and keeping the pixel color of each small subregion uniform as much as possible; establishing an initial color label graph;
the third step: extracting a rectangular frame containing a target by a target detection method, performing color space conversion in the rectangular frame, and converting a color space formed by a RGB color system of a three-color channel into a Lab pixel space;
the fourth step: calculating the color grade of each small sub-region color block, respectively counting the average value of pixels in each sub-region, performing Euclidean distance calculation with the 25 calibrated color labels, and selecting the color label with the minimum Euclidean distance as the color label corresponding to the current sub-region;
the fifth step: setting: the outermost element of the color label graph is a background; starting from each background element on the outermost periphery, searching from the boundary to the inside by searching for the connectivity between the elements, and expanding the subareas one by one until all connected domains connected with the background elements are searched;
a sixth step: eliminating all searched connected domains to obtain a color label map only containing foreground information;
a seventh step of: eliminating all background information to obtain an image with prominent foreground information;
an eighth step: performing AND operation on the image with the outstanding foreground information and the original image picture to obtain an image with the background image removed as a result;
a ninth step: and outputting the image without the background image as a result, and performing target tracking initialization and feature extraction. The extracted features only contain target information, so that background interference is reduced, and the target tracking performance under a complex background can be quickly and effectively improved.
Claims (4)
1. A target tracking rapid initialization method based on color labels is characterized in that,
the first step is as follows: dividing colors into N Lab color grades in a picture, wherein each color grade is assigned with a label numerical value, and the color labels are 1-N in total, namely divided into N colors;
the second step is as follows: dividing the image into subareas, vertically and horizontally cutting one image into a plurality of small subareas, wherein the size of each subarea is M x M, and the pixel color of each small subarea is kept uniform;
the third step: extracting a rectangular frame containing a target by a target detection method, performing color space conversion in the rectangular frame, and converting a color space formed by a RGB color system of a three-color channel into a Lab pixel space;
the fourth step: calculating the color grade of each small sub-region color block, respectively counting the average value of pixels in each sub-region, performing Euclidean distance calculation with the N calibrated color labels, and selecting the color label with the minimum Euclidean distance as the color label corresponding to the current sub-region;
the fifth step: setting: the outermost element of the color label graph is a background; starting from each background element on the outermost periphery, searching from the boundary to the center of the picture by searching for the connectivity between the elements, and expanding the subareas one by one until all connected domains connected with the background elements are searched;
a sixth step: eliminating all searched connected regions to obtain a color label map only containing a foreground information region;
a seventh step of: eliminating all background information areas except the foreground information area to obtain an image with prominent foreground information;
an eighth step: performing AND operation on the image with the outstanding foreground information obtained after eliminating all background information areas except the foreground information and the original image picture to obtain the original foreground information image after removing the background image;
a ninth step: and outputting the foreground information image without the background image as a result, and performing target tracking initialization and feature extraction.
2. The color label-based object tracking fast initialization method according to claim 1, wherein colors in a picture are divided into N Lab color levels, where N is 15-1024.
3. The color label-based target tracking fast initialization method according to claim 1, wherein the frame is cut into small sub-regions vertically and horizontally, each sub-region having a size of M × M, where M is 1-10 pixels.
4. The method as claimed in claim 1, wherein the outermost element of the color label graph is set as a background, a connected domain connected to the background element is found by searching and expanding from outside to inside, and the connected domain is removed to obtain an image containing only the target information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711334249.6A CN109003289B (en) | 2017-12-11 | 2017-12-11 | Target tracking rapid initialization method based on color label |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711334249.6A CN109003289B (en) | 2017-12-11 | 2017-12-11 | Target tracking rapid initialization method based on color label |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109003289A CN109003289A (en) | 2018-12-14 |
| CN109003289B true CN109003289B (en) | 2021-04-30 |
Family
ID=64574086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711334249.6A Active CN109003289B (en) | 2017-12-11 | 2017-12-11 | Target tracking rapid initialization method based on color label |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109003289B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188743A (en) * | 2007-09-17 | 2008-05-28 | 深圳先进技术研究院 | An intelligent digital system based on video and its processing method |
| WO2012078026A1 (en) * | 2010-12-10 | 2012-06-14 | Mimos Berhad | Method for color classification and applications of the same |
| CN104899586A (en) * | 2014-03-03 | 2015-09-09 | 阿里巴巴集团控股有限公司 | Method for recognizing character contents included in image and device thereof |
| CN105631455A (en) * | 2014-10-27 | 2016-06-01 | 阿里巴巴集团控股有限公司 | Image main body extraction method and system |
| CN106570488A (en) * | 2016-11-10 | 2017-04-19 | 江苏信息职业技术学院 | Wavelet algorithm based vehicle tracking recognition method |
| CN106874840A (en) * | 2016-12-30 | 2017-06-20 | 东软集团股份有限公司 | Vehicle information recognition method and device |
| CN107424159A (en) * | 2017-07-28 | 2017-12-01 | 西安电子科技大学 | Image, semantic dividing method based on super-pixel edge and full convolutional network |
-
2017
- 2017-12-11 CN CN201711334249.6A patent/CN109003289B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188743A (en) * | 2007-09-17 | 2008-05-28 | 深圳先进技术研究院 | An intelligent digital system based on video and its processing method |
| WO2012078026A1 (en) * | 2010-12-10 | 2012-06-14 | Mimos Berhad | Method for color classification and applications of the same |
| CN104899586A (en) * | 2014-03-03 | 2015-09-09 | 阿里巴巴集团控股有限公司 | Method for recognizing character contents included in image and device thereof |
| CN105631455A (en) * | 2014-10-27 | 2016-06-01 | 阿里巴巴集团控股有限公司 | Image main body extraction method and system |
| CN106570488A (en) * | 2016-11-10 | 2017-04-19 | 江苏信息职业技术学院 | Wavelet algorithm based vehicle tracking recognition method |
| CN106874840A (en) * | 2016-12-30 | 2017-06-20 | 东软集团股份有限公司 | Vehicle information recognition method and device |
| CN107424159A (en) * | 2017-07-28 | 2017-12-01 | 西安电子科技大学 | Image, semantic dividing method based on super-pixel edge and full convolutional network |
Non-Patent Citations (2)
| Title |
|---|
| "基于Lab颜色空间的运动目标检测";许莉 等;《华中科技大学学报（自然科学版）》;20131031;第41卷;全文 * |
| "面向语义的图像中主要对象的提取方法";孙超 等;《电视技术》;20131231(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109003289A (en) | 2018-12-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Babu et al. | Vehicle number plate detection and recognition using bounding box method | |
| Neumann et al. | Efficient scene text localization and recognition with local character refinement | |
| Ruta et al. | Real-time traffic sign recognition from video by class-specific discriminative features | |
| TWI281126B (en) | Image detection method based on region | |
| CN107392141B (en) | Airport extraction method based on significance detection and LSD (least squares distortion) line detection | |
| CN106650553A (en) | License plate recognition method and system | |
| CN106875546A (en) | A kind of recognition methods of VAT invoice | |
| CN106778757A (en) | Scene text detection method based on text conspicuousness | |
| CN101551853A (en) | Human ear detection method under complex static color background | |
| CN106157323A (en) | The insulator division and extracting method that a kind of dynamic division threshold value and block search combine | |
| CN110008900B (en) | Method for extracting candidate target from visible light remote sensing image from region to target | |
| Balaji et al. | Smart vehicle number plate detection system for different countries using an improved segmentation method | |
| Shivakumara et al. | Gradient-angular-features for word-wise video script identification | |
| CN108537816A (en) | A kind of obvious object dividing method connecting priori with background based on super-pixel | |
| CN106022223A (en) | High-dimensional local-binary-pattern face identification algorithm and system | |
| CN106845458A (en) | A kind of rapid transit label detection method of the learning machine that transfinited based on core | |
| CN108073925A (en) | A kind of licence plate recognition method and device | |
| Song et al. | Depth-aware saliency detection using discriminative saliency fusion | |
| TWI628623B (en) | All-weather thermal image type pedestrian detection method | |
| CN105354547A (en) | Pedestrian detection method in combination of texture and color features | |
| CN101488182B (en) | Image characteristics extraction method used for handwritten Chinese character recognition | |
| CN107679467B (en) | Pedestrian re-identification algorithm implementation method based on HSV and SDALF | |
| Giri | Text information extraction and analysis from images using digital image processing techniques | |
| CN109003289B (en) | Target tracking rapid initialization method based on color label | |
| Zhang et al. | A novel approach for binarization of overlay text |
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 | ||
| CB02 | Change of applicant information |
Address after: 361008 Fujian province Xiamen Software Park Siming District Road No. 59 102 two expected Applicant after: ROPT TECHNOLOGY GROUP Co.,Ltd. Address before: 361008 Fujian province Xiamen Software Park Siming District Road No. 59 102 two expected Applicant before: Roput (Xiamen) Technology Group Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |