CN105139417A - Method for real-time multi-target tracking under video surveillance - Google Patents
Method for real-time multi-target tracking under video surveillance Download PDFInfo
- Publication number
- CN105139417A CN105139417A CN201510446292.6A CN201510446292A CN105139417A CN 105139417 A CN105139417 A CN 105139417A CN 201510446292 A CN201510446292 A CN 201510446292A CN 105139417 A CN105139417 A CN 105139417A
- Authority
- CN
- China
- Prior art keywords
- target
- similarity
- color
- template
- real time
- 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.)
- Pending
Links
Classifications
-
- 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
- G06T2207/10016—Video; Image sequence
Landscapes
- Image Analysis (AREA)
Abstract
The present invention discloses a method for real-time multi-target tracking under video surveillance, and belongs to the technical field of image processing. The method comprises: screening foreground regions obtained by pre-processing each frame of input image to obtain a target region; obtaining a color block edge direction histogram and a HSI color histogram; and in the process of calculating the similarity of a candidate target and a target template within a target search range and obtaining a tracking result according to the similarity, after describing the similarity of the color block edge direction histogram and the similarity of the HSI color histogram, assigning weight values to the similarity of the color block edge direction histogram and the similarity of the HSI color histogram to determine the similarity of the candidate target and the target template, wherein the target template comprises all targets of a previous frame of input image within the target search range. According to the method, the accuracy and robustness are improved; and a parallelizing process is performed on a matching process, so that the time required to process each frame of image is greatly shortened and real-time requirements are met.
Description
Technical field
The invention discloses the multiple target method for real time tracking under video monitoring, belong to the technical field of image procossing.
Background technology
Human body behavioural analysis in video is a study hotspot in recent years, and the bottom vision technique such as human motion detect and track is to realize Human bodys' response and understand the basis of contour level vision technique.In existing target following research, commonly used target property mainly includes the visual signatures such as edge, profile of object and the statistical nature counted to color category.Color characteristic is insensitive with the rotation to target and local deformation, the characteristics of robustness is good, and many advantages such as the dimensional variation of target can be adapted to a certain extent, it is used widely in terms of target following, but in illumination, color change and when having Similar color interference, deviation occurs in tracking result.The advantages of marginal information is with being influenceed small by illumination variation, and not needing specific object module, but human body is nonrigid, edge, contour feature can be continually changing, so the human body tracking poor robustness based on edge, accuracy rate is low.The method for being combined color characteristic and edge feature was it is also proposed in the last few years, although improving target discrimination, but objective contour is changed greatly when occurring staggeredly to block, influence the accuracy of matching result, it the method increase accuracy rate and stability, but it is due to that to be related to feature more, complexity is high, it is difficult to meet the requirement of real-time.
At present, intelligent monitoring is for real-time and the requirement more and more higher of accuracy, and conventional method is difficult to the demand for meeting current practical application.
The content of the invention
The technical problems to be solved by the invention are the deficiencies for above-mentioned background technology, there is provided the multiple target method for real time tracking under video monitoring, color and two kinds of edge Feature Descriptor is considered to calculate the similarity of front and rear two field pictures, by giving color lump edge histogram similarity, HSI color histograms similarity assigns weights to describe the similarity of target and former frame target, improve accuracy rate and robustness, color characteristic is solved in illumination, color change and having can cause tracking result deviation occur during Similar color interference, human body tracking poor robustness based on edge, color still technical problem of matching precision difference when target is interlocked and blocked with tracking that edge is combined.
The present invention is adopted the following technical scheme that for achieving the above object：
Multiple target method for real time tracking under video monitoring, comprises the following steps：
The foreground area obtained to pre-processing each frame input picture screen obtaining target area；
Obtain color lump edge orientation histogram and HSI color histograms；
The similarity of candidate target and To Template is calculated in the range of target search and tracking result is obtained by similarity：Describe after color lump edge orientation histogram similarity, HSI color histogram similarities, weights are assigned to determine the similarity of candidate target and To Template to color lump edge orientation histogram similarity, HSI color histograms similarity, and the To Template includes former frame input picture all targets in target search scope.
Further, in the multiple target method for real time tracking under the video monitoring, the method for obtaining color lump edge orientation histogram is：
Use level and two vertical Sobel operators calculate the horizontal gradient and vertical gradient of each pixel in target area, then obtain the gradient direction of each pixel by horizontal gradient and vertical gradient, and each pixel gradient direction is quantified to obtain color lump edge orientation histogram.
Further, in the multiple target method for real time tracking under the video monitoring, target search scope is determined by kalman filter method, is specially：According to the position of target, speed and direction in previous frame input picture, the position range that target occurs in present frame input picture is determined using Kalman filtering.
Further, multiple target method for real time tracking under the video monitoring, in the step for similarity of candidate target and To Template is calculated in the range of target search and tracking result is obtained by similarity, using Pasteur distance description color lump edge orientation histogram similarity BH, HSI color histogram similarity BG, according to expression formula：R=0.7BH+0.3BG assigns weights to determine the similarity R of candidate target and To Template to color lump edge orientation histogram similarity, HSI color histograms similarity.
Further, multiple target method for real time tracking under the video monitoring, the similarity of candidate target and To Template is calculated in the range of target search and is completed parallel with target numbers identical thread by GPU the step for obtaining tracking result by similarity, each thread calculates its correspondence target and the similarity of To Template, each thread output result that sorts obtains optimal result, when optimal result is more than or equal to threshold value, the match is successful, and when optimal result is still less than threshold value, it fails to match.
Further, in the multiple target method for real time tracking under the video monitoring, foreground area is pre-processed using background subtraction method to input picture.
Further, the multiple target method for real time tracking under the video monitoring, fresh target is screened for the human body target in input picture by surrounding the rectangle size of foreground area.
Further, the multiple target method for real time tracking under the video monitoring, target area is screened by fresh target in foreground area with other existing target ranges.
The present invention uses above-mentioned technical proposal, has the advantages that：Color characteristic and color lump edge feature are combined, with the similitude for the Pasteur's distance description image for considering color characteristic and color lump edge feature, improve accuracy rate and robustness, and parallelization processing is carried out to matching process, time of the processing required for per two field picture is greatly reduced, the requirement of real-time has been reached.
The additional aspect of the present invention and advantage will be set forth in part in the description, and these will become apparent from the description below, or be recognized by the practice of the present invention.
Brief description of the drawings
Fig. 1 is the present invention relates to the flow chart of tracking.
Fig. 2 (a) to Fig. 2 (d) be respectively traditional color tracking algorithm to video real-time tracking when the 70th frame, the 78th frame, the 85th frame, the experiment effect figure of the 110th frame.
Fig. 3 (a) to Fig. 3 (d) be respectively traditional border following algorithm to video real-time tracking when the 70th frame, the 78th frame, the 85th frame, the experiment effect figure of the 110th frame.
The 70th frame, the 78th frame, the 85th frame, the experiment effect figure of the 110th frame that Fig. 4 (a) to Fig. 4 (d) is respectively color when track algorithm that edge feature is combined is to video real-time tracking.
Fig. 5 (a) to Fig. 5 (d) be respectively the present invention to video real-time tracking when the 70th frame, the 78th frame, the 85th frame, the experiment effect figure of the 110th frame.
Embodiment
Embodiments of the present invention are described below in detail, the embodiment described below with reference to accompanying drawing is exemplary, are only used for explaining the present invention, and be not construed as limiting the claims.
Multiple target method for real time tracking under video monitoring of the present invention is as shown in Figure 1, color and two kinds of edge Feature Descriptor are considered to calculate the similarity of front and rear two field pictures, by describing the similarity of target and former frame target to color lump edge histogram similarity, HSI color histograms similarity imparting weights.
Step A, input picture is carried out to pre-process and obtain foreground area, then each foreground area screen obtaining fresh target region：Foreground area is obtained first by background subtraction method, then foreground area is surrounded with rectangle, human body is determined whether by rectangle size, whether is fresh target finally by the Distance Judgment of the target and existing target.
Step B, utilize following methods extract target signature：
Step B1, the color lump edge orientation histogram for obtaining using following methods target：
Step B101, for the target area in image, use level and two vertical Sobel operators calculate pixel p in target areai,jThe horizontal gradient Gx of (wherein, i, j represent the line number and columns where pixel)i,jWith vertical gradient Gyi,j, so as to calculate pixel pi,jGradient direction Dir (pi,j),
Step B102, by Dir (pi,j) span be divided into n equal portions so that by Dir (pi,j) quantified, it is assumed thatWherein, k=0,1,2,3, then to Dir (pi,j) value N (p after quantizationi,j)=k.According to N (pi,j) just obtain color lump edge histogram H；
Step B2, HSI (HueSaturationIntensity) color histogram for obtaining using following methods target：
Step B201, the RGB image of input is converted into HSI images,
Step B202, by HSI color spaces non-uniform quantizing into 72 dimensions, tone H is divided into 8 parts, and saturation degree S and brightness I are respectively divided into 3 parts, if G=9H+3S+I, therefore obtain 72 handle histogram G.
Step C, target search scope determined by the method for Kalman filtering, specifically include following steps：
Step C1, target location, speed and the direction obtained in previous frame result.
Step C2, the position range that target is likely to occur in present frame input picture is obtained by Kalman filtering.
Step D, using GPU to matching process carry out parallelization processing, quickly obtain in the best match position of each target, present embodiment using NVDIA release CUDA computing platforms, by GPU handle object matching process；This step specifically includes following steps：
Step D1, the Thread Count for determining to need according to destination number, match information is imported in GPU；
Step D2, in each thread, according to the obtained target search scope of thread number and step C determine the thread be used for match target and candidate target corresponding position in the picture；
Step D3, the method extraction candidate target feature using step B, use Pasteur's distance description candidate target and the color lump edge orientation histogram and the similarity of color histogram of To Template, color lump edge orientation histogram is obtained apart from BH and color histogram map distance BG, then assign BG the weights different with BH, last similarity R is obtained as output result according to formula R=0.7BH+0.3BG；
After step D4, parallel processing are finished, the output result of all threads is ranked up to the best matching result for obtaining each target, if optimal result is still less than Pasteur's distance threshold 0.6, then it is assumed that it fails to match；
Step D5, show matching result and update each target information.
Experiment is tracked present invention uses many people's walking videos in one section of interior, video length is 200 frames, video image resolution ratio 640*480.Traditional color tracking algorithm is for shown in experiment effect such as Fig. 2 (a) to Fig. 2 (d) of the video, during tracking, because color of object is close, can frequently occur target misrecognition, and accuracy rate is very low.Traditional border following algorithm is for shown in experiment effect such as Fig. 3 (a) to Fig. 3 (d) of the video, when the overlapping combination of target, profile is varied widely, it is impossible to be tracked identification.The track algorithm that color is combined with edge feature is for shown in experiment effect such as Fig. 4 (a) to Fig. 4 (d) of the video, due to combining color characteristic and edge feature, improve the discrimination between target, but when occurring staggeredly to block, wherein profile variations are larger, matching result is influenceed, the raising to accuracy rate is not that very big and efficiency is substantially reduced.Shown in the effect of inventive algorithm such as Fig. 5 (a) to Fig. 5 (d), due to adding the statistics to color lump edge, improve the discrimination between target, it can also be tracked well when occurring and staggeredly blocking, improve the accuracy rate of tracking, and parallelization processing has been carried out to the algorithm, the requirement of real-time can be reached.Efficiency comparative's figure of four kinds of algorithms is as shown in table 1.
1 four kinds of track algorithm efficiency comparatives of table
It can be seen that, color characteristic and color lump edge feature are combined by the present invention, improve accuracy rate and robustness, and carry out parallelization processing to matching process, can be accomplished 1 second processing 10-15 two field picture under conditions of external environment is suitable, reached the requirement of real-time.This method can be applied to the fields such as bank monitoring, the efficiency of monitoring system can be also significantly increased while saving human cost.
In summary, the invention has the advantages that：Color characteristic and color lump edge feature are combined, with the similitude for the Pasteur's distance description image for considering color characteristic and color lump edge feature, improve accuracy rate and robustness, and parallelization processing is carried out to matching process, time of the processing required for per two field picture is greatly reduced, the requirement of real-time has been reached.
One of ordinary skill in the art will appreciate that：Accompanying drawing is necessary to module or flow in the schematic diagram of one embodiment, accompanying drawing not necessarily implements the present invention.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can add the mode of required general hardware platform to realize by software.Understood based on such, the part that technical scheme substantially contributes to prior art in other words can be embodied in the form of software product, the computer software product can be stored in storage medium, such as ROM/RAM, magnetic disc, CD, including some instructions to cause a computer equipment (can be personal computer, server, or network equipment etc.) to perform the method described in some parts of embodiments of the invention or embodiment.
Claims (8)
1. the multiple target method for real time tracking under video monitoring, it is characterised in that comprise the following steps：
The foreground area obtained to pre-processing each frame input picture screen obtaining target area；
Obtain color lump edge orientation histogram and HSI color histograms；
The similarity of candidate target and To Template is calculated in the range of target search and tracking result is obtained by similarity：Describe after color lump edge orientation histogram similarity, HSI color histogram similarities, weights are assigned to determine the similarity of candidate target and To Template to color lump edge orientation histogram similarity, HSI color histograms similarity, and the To Template includes former frame input picture all targets in target search scope.
2. the multiple target method for real time tracking under video monitoring according to claim 1, it is characterised in that the method for obtaining color lump edge orientation histogram is：Use level and two vertical Sobel operators calculate the horizontal gradient and vertical gradient of each pixel in target area, then obtain the gradient direction of each pixel by horizontal gradient and vertical gradient, and each pixel gradient direction is quantified to obtain color lump edge orientation histogram.
3. the multiple target method for real time tracking under video monitoring according to claim 1, it is characterised in that the target search scope is determined by kalman filter method, is specially：According to the position of target, speed and direction in previous frame input picture, the position range that target occurs in present frame input picture is determined using Kalman filtering.
4. the multiple target method for real time tracking under video monitoring according to claim 1, it is characterized in that, in the step for similarity of candidate target and To Template is calculated in the range of target search and tracking result is obtained by similarity, using Pasteur distance description color lump edge orientation histogram similarity BH, HSI color histogram similarity BG, according to expression formula：R=0.7BH+0.3BG assigns weights to determine the similarity R of candidate target and To Template to color lump edge orientation histogram similarity, HSI color histograms similarity.
5. the multiple target method for real time tracking under video monitoring according to claim 1 or 2 or 3 or 4, it is characterized in that, the similarity of candidate target and To Template is calculated in the range of target search and is completed parallel with target numbers identical thread by GPU the step for obtaining tracking result by similarity, each thread calculates its correspondence target and the similarity of To Template, each thread output result that sorts obtains optimal result, when optimal result is more than or equal to threshold value, the match is successful, and when optimal result is still less than threshold value, it fails to match.
6. the multiple target method for real time tracking under video monitoring according to claim 1, it is characterised in that the foreground area is pre-processed using background subtraction method to input picture.
7. the multiple target method for real time tracking under video monitoring according to claim 1, it is characterised in that screen fresh target by surrounding the rectangle size of foreground area for the human body target in input picture.
8. the multiple target method for real time tracking under video monitoring according to claim 7, it is characterised in that target area is screened with other existing target ranges by fresh target in foreground area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510446292.6A CN105139417A (en) | 2015-07-27 | 2015-07-27 | Method for real-time multi-target tracking under video surveillance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510446292.6A CN105139417A (en) | 2015-07-27 | 2015-07-27 | Method for real-time multi-target tracking under video surveillance |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105139417A true CN105139417A (en) | 2015-12-09 |
Family
ID=54724749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510446292.6A Pending CN105139417A (en) | 2015-07-27 | 2015-07-27 | Method for real-time multi-target tracking under video surveillance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105139417A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106097385A (en) * | 2016-05-31 | 2016-11-09 | 海信集团有限公司 | A kind of method and apparatus of target following |
| CN106295532A (en) * | 2016-08-01 | 2017-01-04 | 河海大学 | A kind of human motion recognition method in video image |
| CN106485733A (en) * | 2016-09-22 | 2017-03-08 | 电子科技大学 | A kind of method following the tracks of interesting target in infrared image |
| CN107301651A (en) * | 2016-04-13 | 2017-10-27 | 索尼公司 | Object tracking apparatus and method |
| CN109446901A (en) * | 2018-09-21 | 2019-03-08 | 北京晶品特装科技有限责任公司 | A kind of real-time humanoid Motion parameters algorithm of embedded type transplanted |
| CN109495749A (en) * | 2018-12-24 | 2019-03-19 | 上海国茂数字技术有限公司 | A kind of coding and decoding video, search method and device |
| CN111950218A (en) * | 2020-07-02 | 2020-11-17 | 深圳市兴森快捷电路科技股份有限公司 | Circuit for realizing target tracking algorithm based on FPGA |
| CN111982120A (en) * | 2020-08-20 | 2020-11-24 | 青岛海米飞驰智能科技有限公司 | Search and rescue target identification method, system and equipment based on water robot |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103065331A (en) * | 2013-01-15 | 2013-04-24 | 南京工程学院 | Target tracking method based on correlation of space-time-domain edge and color feature |
-
2015
- 2015-07-27 CN CN201510446292.6A patent/CN105139417A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103065331A (en) * | 2013-01-15 | 2013-04-24 | 南京工程学院 | Target tracking method based on correlation of space-time-domain edge and color feature |
Non-Patent Citations (4)
| Title |
|---|
| DORIN COMANICIU ET AL: "Kernel-Based Object Tracking", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| YUNQIANG CHEN ET AL: "JPDAF Based HMM for Real-Time Contour Tracking", 《CVPR 2001》 * |
| 刘惟锦 等: "基于Kalman滤波和边缘直方图的实时目标跟踪", 《清华大学学报(自然科学版)》 * |
| 王绍钰 等: "改进的粒子滤波器目标跟踪方法", 《智能系统学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107301651A (en) * | 2016-04-13 | 2017-10-27 | 索尼公司 | Object tracking apparatus and method |
| CN106097385A (en) * | 2016-05-31 | 2016-11-09 | 海信集团有限公司 | A kind of method and apparatus of target following |
| CN106097385B (en) * | 2016-05-31 | 2019-03-05 | 海信集团有限公司 | A kind of method and apparatus of target following |
| CN106295532A (en) * | 2016-08-01 | 2017-01-04 | 河海大学 | A kind of human motion recognition method in video image |
| CN106295532B (en) * | 2016-08-01 | 2019-09-24 | 河海大学 | A kind of human motion recognition method in video image |
| CN106485733A (en) * | 2016-09-22 | 2017-03-08 | 电子科技大学 | A kind of method following the tracks of interesting target in infrared image |
| CN109446901A (en) * | 2018-09-21 | 2019-03-08 | 北京晶品特装科技有限责任公司 | A kind of real-time humanoid Motion parameters algorithm of embedded type transplanted |
| CN109495749A (en) * | 2018-12-24 | 2019-03-19 | 上海国茂数字技术有限公司 | A kind of coding and decoding video, search method and device |
| CN111950218A (en) * | 2020-07-02 | 2020-11-17 | 深圳市兴森快捷电路科技股份有限公司 | Circuit for realizing target tracking algorithm based on FPGA |
| CN111982120A (en) * | 2020-08-20 | 2020-11-24 | 青岛海米飞驰智能科技有限公司 | Search and rescue target identification method, system and equipment based on water robot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105139417A (en) | Method for real-time multi-target tracking under video surveillance | |
| CN108549864B (en) | Vehicle-mounted thermal imaging pedestrian detection-oriented region-of-interest filtering method and device | |
| CN108549874B (en) | Target detection method, target detection equipment and computer-readable storage medium | |
| CN109903331B (en) | Convolutional neural network target detection method based on RGB-D camera | |
| CN104408429B (en) | A kind of video represents frame extracting method and device | |
| Yan et al. | One extended OTSU flame image recognition method using RGBL and stripe segmentation | |
| CN110688987A (en) | Pedestrian position detection and tracking method and system | |
| CN107316031A (en) | The image characteristic extracting method recognized again for pedestrian | |
| CN107705288A (en) | Hazardous gas spillage infrared video detection method under pseudo- target fast-moving strong interferers | |
| CN107358245B (en) | Method for detecting image collaborative salient region | |
| CN105809716B (en) | Foreground extraction method integrating superpixel and three-dimensional self-organizing background subtraction method | |
| CN104966305B (en) | Foreground detection method based on motion vector division | |
| CN105740945A (en) | People counting method based on video analysis | |
| WO2017166597A1 (en) | Cartoon video recognition method and apparatus, and electronic device | |
| CN106600613B (en) | Improvement LBP infrared target detection method based on embedded gpu | |
| CN106529441A (en) | Fuzzy boundary fragmentation-based depth motion map human body action recognition method | |
| US9940543B2 (en) | Control of computer vision pre-processing based on image matching using structural similarity | |
| CN103578121B (en) | Method for testing motion based on shared Gauss model under disturbed motion environment | |
| CN113034383B (en) | Method for acquiring video image based on improved grid motion statistics | |
| CN107977608B (en) | Method for extracting road area of highway video image | |
| CN111461002B (en) | Sample processing method for thermal imaging pedestrian detection | |
| CN111626107B (en) | Humanoid contour analysis and extraction method oriented to smart home scene | |
| CN106446832B (en) | Video-based pedestrian real-time detection method | |
| CN111161219B (en) | Robust monocular vision SLAM method suitable for shadow environment | |
| CN102129569A (en) | Equipment and method for detecting object based on multiscale comparison characteristic |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20151209 |
|
| WD01 | Invention patent application deemed withdrawn after publication |