CN101867699A - Real-time tracking method of nonspecific target based on partitioning - Google Patents

Real-time tracking method of nonspecific target based on partitioning Download PDF

Info

Publication number
CN101867699A
CN101867699A CN201010184876A CN201010184876A CN101867699A CN 101867699 A CN101867699 A CN 101867699A CN 201010184876 A CN201010184876 A CN 201010184876A CN 201010184876 A CN201010184876 A CN 201010184876A CN 101867699 A CN101867699 A CN 101867699A
Authority
CN
China
Prior art keywords
target
grader
real
updating
weight
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201010184876A
Other languages
Chinese (zh)
Other versions
CN101867699B (en
Inventor
俞能海
周维
庄连生
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Science and Technology of China USTC
Original Assignee
University of Science and Technology of China USTC
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN2010101848768A priority Critical patent/CN101867699B/en
Publication of CN101867699A publication Critical patent/CN101867699A/en
Application granted granted Critical
Publication of CN101867699B publication Critical patent/CN101867699B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention relates to a real-time tracking method of a nonspecific target based on partitioning, comprising three steps of classifier updating, target detection and weight updating. In the method, the target region is divided into multiple blocks; a classifier is used for maintaining each block, and updating is conducted frame by frame; the detection result of each classifier is comprehensively considered to determine the position of the target in the new video frame. In the method, an automatic weight updating mechanism is designed, so as to enable the blocks which are relatively stable to have greater decision-making power over the judgment of results, thus reducing the influence of various interferences; and the tracking performance is better than multiple international published algorithms recently. In the method, changes of appearances of objects caused by various interferences can be captured and accurate tracking can be conducted; the method has universality on target objects in various shapes and types; the calculation has low complexity and can be processed at real time. The invention has wide application prospect in various occasions needing tracking techniques, such as video monitoring, automatic driving, man-machine interaction, intelligent traffic, robot, airborne early warning and the like.

Description

Real-time tracking method of nonspecific target based on piecemeal
Technical field
The present invention relates to method for tracking target in the video, particularly a kind of real-time tracking method of nonspecific target based on piecemeal.
Background technology
Target following is computer vision and important work of automatic field, all has a very wide range of applications aspect military and civilian.As man-machine interaction, intelligent transportation, security monitoring, robot, air-borne early warning etc.These are used all target following have in real time automatically been proposed higher demand and requirement.
At present video target tracking method has many kinds, as based on profile or template matches, based on filter, based on classification or the like.Relatively more outstanding is the mean-shift track algorithm that was proposed by Dorin Comaniciu in nearest 2 years, (supervision semi-supervised) online Adaboost track algorithm (OABTracker) that the tracking based on fragment (Frag Tracker) that AmitAdam proposes, Helmut Grabner propose and many learn-by-examples track algorithm (MILTracker) of the Boris Babenko of California, USA university proposition.Preceding two kinds of methods all to target travel and the significant target following ability of cosmetic variation a little less than, online Adaboost method can adapt to the cosmetic variation of target preferably, but drift appears easily, the long slightly then possibility of tracking time lose objects, and to the disposal ability of blocking a little less than, the 4th kind of more online Adaboost method of method strengthened the adaptive ability to target appearance, and certain disposal ability of blocking is also arranged, but computational complexity is higher, and real-time is low.These methods all are that target is done as a whole the tracking in addition, can not distinguish ground processing target different piece, the local appearance variation meeting that is caused by attitude, factor such as block produces significant adverse to tracking results influence.People such as Bo Wu proposed branch's formula pedestrian tracting method in 2007, this method is to be the tracking of target at the pedestrian specially, and its piecemeal is structurized head with semantic probability, trunk or the like, need the parts of body detector of training in advance, general target is not had universality.
Summary of the invention
The objective of the invention is to concentrate on target area more stable part relatively with what follow the tracks of, and mainly make judgement according to these parts, adaptive ability and robustness that raising changes target appearance to blocking, remedy the shortcoming of existing method for tracking target, realize tracking accurately and real-time nonspecific type target.Therefore we carry out non-semantic piecemeal processing to target, be that target is carried out grader study and detection with each piece respectively earlier, the result of comprehensive each piece determines the position of target in new video frame then, for each object block is treated with a certain discrimination, we give each piece a weight, and designed a weight update mechanism, upgrade weight according to the tracking results frame by frame of former frame.This method has improved the accuracy of following the tracks of significantly when keeping low computation complexity.
For achieving the above object, the invention provides a kind of real-time tracking method of nonspecific target based on piecemeal, the video frame by frame is carried out target localization, the processing of each frame is comprised that all grader upgrades, target detection and weight are upgraded three big steps.
Described grader learning procedure is:
Step a, piecemeal.According to target location that has obtained and size information, the target area is divided into the rectangular block of even big or small non-overlapping copies, they do not need to have semantic feature, cut apart very simply arbitrary target all is suitable for, and each other related of piece and piece also only is the space syntople.Determine the size and the number of piece when first frame, each follow-up frame is all divided like this, and each piece center and target's center have a fixing side-play amount
Figure GSA00000130688300021
The size of piece is controlled between the 16*16-50*50 pixel, and the number K of piece is controlled between the 4-12, but above restriction is all non-imposed;
Step b safeguards a grader C to each target image piece k, and upgrade online Adaboost grader at each frame, concrete step can be divided into again:
(1) extracts training sample.With current image block is positive sample, and taking out 20-50 image block at random near the zone it is negative sample, extracts N dimension haar-like feature;
(2) upgrade Weak Classifier with online Adaboost algorithm and select n the new C of Weak Classifier composition k
Described target detection step is:
Step c for the t frame that newly obtains, uses grader C kNear the k piece, travel through detection in R pixel radius region, calculate grader to each position (i, j) score value C k(i j) (does not carry out the successive value that classification quantizes), obtains grader shot chart I k, and a further calculating K local confidence map
Figure GSA00000130688300022
p k ( i , j ) = 1 1 + exp ( - 2 I k ( i , j ) ) ;
Steps d obtains overall confidence map P to whole K local confidence map weighted average:
Figure GSA00000130688300024
It is the side-play amount that the piece central point is put to target's center, the relativity shift of each piece has been considered in the weighted sum of portion's confidence map of playing a game, the overall confidence level of trying to achieve is P (i, what j) express is that whole target is in position (i, j) confidence, making does not so need to consider the relative position relation of piece and target in step c;
Step e adopts the mean-shift method to seek on overall confidence map and puts the letter peak value, this position both be considered to the position pos=of target in the t frame (x, y)=meanshift (P);
Described weight step of updating is:
Step f is to each piece shot chart I k| I k(i, j)=C k(i j), thinks at pixel pos+ (x k, y k) near r 1In the individual pixel coverage is the positive feedback value, a Gaussian Profile N of match (μ 1, σ 1), r 2That individual pixel coverage is outer is negative feedback value, also a Gaussian Profile N of match (μ 0, σ 0), calculate classification thresholds thus
Figure GSA00000130688300025
This is theoretical minimal error circle that two Gausses distribute, as Fig. 1;
Step g is used T kAlign the negative feedback value and classify, the mistake in computation rate
Figure GSA00000130688300031
And the confidence level of calculating grader λ k t = log ( ( 1 - e k t ) / e k t ) ;
Step h upgrades the weight of each piece
Figure GSA00000130688300033
Be hysteresis factors, α is usually near 1, and is steady to guarantee that weight changes, and improves the stability of a system; And the grader update mark is set according to the error rate size of each piece
Figure GSA00000130688300035
If
Figure GSA00000130688300036
Grader C is described kError rate at present frame is too high, and this moment, grader can not resolution target and background, and whether then ensuing grader or not grader C when upgrading k, suspend grader and upgrade, can avoid the introducing of more noises.
Beneficial effect of the present invention is, by piecemeal, the piece of different conditions is treated with a certain discrimination.Mainly be renewal by weight, the piece that cosmetic variation is violent or occurrence of large-area is blocked, the online Adaboost grader C of its correspondence kResolution target and background preferably, this method just reduce its right to speak automatically, reduce the interference that the target location is judged, opposite metastable then can obtain increasing weight, and it will increase tribute of judging, and the result of tracking will be more accurate.Method of the present invention without any need for study in advance, can both handle and can real-time tracking the target of any type.
Experimental result shows that the present invention inherits also and has significantly improved online Adaboost method to illumination, attitude, and motion such as blocks at the robustness that caused target appearance changes.As Fig. 2 and table 1.The present invention adopts low feature of computation complexity and learning algorithm in addition, and candidate's Weak Classifier and iterations are few, can realize real-time tracking (when search radius R was 15 pixels, common Pentium Dual Core 2.0G CPU processing speed can reach per second and surpass 30 frames).
The accuracy of table 1 tracking relatively, video data except that Squeezer be take voluntarily, all derive from network http://www.cs.toronto.edu/～dross/ivt/; Http:// vision.ucsd.edu/～bbabenko/project_miltrack.shtml; And http://www.cs.technion.ac.il/～amita/fragtrack/fragtrack.htm, be the frequently-used data of this area research.
Figure GSA00000130688300037
??Caviar?occlusion ??42 ??26 ??8
??Squeezer ??24 ??42 ??10
??Sylvester ??23 ??12 ??20
??Coke?can ??25 ??21 ??14
Description of drawings
Fig. 1 is Gauss curve fitting and classification thresholds schematic diagram.
Fig. 2 is the tracking effect comparison diagram, and left figure video is walking woman, the corresponding video Coke of right figure can.
Fig. 3 is based on the trace flow schematic diagram of piecemeal.
Fig. 4 is a more new technological process of grader.
Fig. 5 is the target detection flow chart.
Fig. 6 is that weight is upgraded flow chart.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment only is a part of embodiment of the present invention, rather than whole embodiment.Based on embodiments of the invention, the every other embodiment that those of ordinary skills are obtained under the prerequisite of not making creative work belongs to the scope of protection of the invention.
As shown in Figure 3, the present invention is based on the nonspecific target real-time tracking flow process intention of piecemeal for application.For an input video, set the goal in the position and the size of first frame, earlier the target area is divided into image block, utilize the grader of each piece correspondence of these Data Update, on subsequent frame, detect then, determine the position that target is new, last result according to detection upgrades weight, so circulation, up to definite target position of a frame in the end, the weight of each piece of initial time is set to equate.What deserves to be mentioned is in first frame, more than one of the positive sample that is used to learn, but side-play amount all is considered as positive sample less than the image block of 3 pixels.
Before the process to tracking is elaborated, thought of the present invention is illustrated.Core concept of the present invention is exactly that the different piece of target is treated with a certain discrimination, focus onto above metastable, with their grader response as the main foundation of judging, to improve the accuracy and the stability of tracking.
The present invention does not have specific (special) requirements to target classification and shape, can effectively follow the tracks of the accuracy height for the target of any classification, and the energy real-time tracking can be widely used in the occasion that various needs are followed the tracks of, for example video monitoring, film is analyzed, man-machine interaction or the like.Other online and incremental learning method can both be applied to framework of the present invention very easily in addition.
Below tracing process is elaborated.
As shown in Figure 4, grader of the present invention more new technological process be:
Step 101: beginning;
Step 102: judge whether first frame, if then according to the big or small S and the number K of target area size estimation piece, S is controlled between the 16*16-50*50 pixel, and the number K of piece is controlled between the 4-12, and the division of subsequent frame is consistent with first frame.Otherwise execution in step 103;
Step 103: piecemeal is carried out in the target area;
Step 104:k=k+1 judges k≤K, is execution in step 105 then, finishes otherwise upgrade;
Step 105: judge
Figure GSA00000130688300051
Be then to get back to step 104;
Step 106: prepare learning sample.Calculate the integrogram of whole image,, and randomly draw (20-40) piece in its vicinity, calculate the N dimension haar-like feature of each sample according to integrogram as negative sample with the positive sample of current block as the renewal grader;
Step 107: carry out grader with online Adaboost algorithm and upgrade, select n Weak Classifier of error rate minimum to form new grader;
Step 108: execution in step 104.
As Fig. 5, the flow process of target detection is:
Step 201:t=t+1 switches to new frame of video;
Step 202: the detection range of determining each grader.Each piece is expanded R pixel up and down on the position of former frame, the rectangular area of formation is as scope to be detected;
Step 203: judge k＞K? whether promptly finish after testing, be execution in step 207 then;
Step 204: use grader C kTraversal detects in the scope to be detected of correspondence, obtains block classifier shot chart I k
Step 205: the grader score is converted into confidence level, obtains K local confidence map:
p k ( i , j ) = 1 1 + exp ( - 2 C k ( i , j ) ) ;
Step 206:k=k+1, execution in step 203;
Step 207: calculate overall confidence map, be about to each local confidence map weighted sum
P ( i , j ) = Σ k = 1 K ω k p k ( i + x k , j + y k )
Step 208: adopt the mean-shift algorithm to estimate the peak of confidence map.Herein promptly as the position of target in new video frame.
As Fig. 6, the flow process that weight is upgraded is:
Step 301: beginning;
Step 302: judge k＞K? being whether confidence level finishes as calculated, is execution in step 308 then;
Step 303: to each shot chart I k, with target location skew (x k, y k) be the center, be r at radius 1(recommended value: what 3) scope was interior must score value be positive feedback, and dropping on radius is r 2(recommended value: 5) outside the scope for negative feedback;
Step 304: the average and the variance (μ that estimate its positive and negative value of feedback respectively 1, σ 1μ 0, σ 0);
Step 305: calculate classification thresholds
Figure GSA00000130688300061
Step 306: align the negative sample collection with this threshold value and classify the error of calculation
Figure GSA00000130688300062
And confidence level λ k t + 1 = log ( ( 1 - e k t + 1 ) / e k t + 1 ) ;
Step 307:k=k+1, execution in step 302;
Step 308: to the confidence level of whole block classifiers
Figure GSA00000130688300064
Carry out normalization;
Step 309: the weight of upgrading each piece Be hysteresis factors, recommended value is 0.75-0.9;
Step 310: reset the update mark of each grader
Figure GSA00000130688300066
E is an error rate threshold, and less than 0.5, recommended value is 0.3-0.4.
The present invention is for unspecific target type, only according to spatial relationship the target area is divided into the piece of non-overlapping copies, gives weight to each piece, and in follow-up tracing process online updating grader and weight, can effectively catch target appearance and change, and make the self adaptation adjustment, focus onto more stable piece, can reduce the interference of destabilizing factor, to illumination, attitude, motion, the variation of multiple factor such as block, all have good robustness.And tracking principle of the present invention is simple, and computation complexity is little, can accomplish real-time processing on ordinary PC.But and have very strong flexibility, and other some learning methods, feature and update method can both be applied in this method easily.Parameter in above in addition the description is all adjustable, and for the processing that the identical parameters of setting can both efficiently and accurately in the tracking of different target is used, does not need other adjustment.Though above description all is the tracking at single target, only need does simple modification and can realize multiobject tracking.
The above description of this invention is illustrative, and it is nonrestrictive, those skilled in the art is understood, within spirit that claim limits and scope, can carry out many modifications to it, change or equivalence, be not limited to online Adaboost as the grader learning method, online MILBoost, many other learning methods such as online SVM can both be used for wherein, the feature representation of image also can be replaced many other features, scale factor for example can be set again to adapt to the variation of target scale in the target detection step, all image zoom is detected to a plurality of yardsticks at every turn, final when determining the target location not only spatially, and on yardstick, get confidence peaks, These distortion all will fall within the scope of protection of the present invention.

Claims (8)

1. based on the real-time tracking method of nonspecific target of piecemeal, comprise that grader upgrades, target detection and weight are upgraded three big steps, it is characterized in that described grader step of updating is:
Step a, piecemeal according to target location that has obtained and size information, is divided into the target area rectangular block of even big or small non-overlapping copies;
Step b safeguards a grader C to each region unit k, and carry out online updating at each frame;
Described target detection step is:
Step c detects on new video frame with grader, and calculates local confidence map p k
Steps d obtains overall confidence map P to whole K local confidence map weighted average;
Step e adopts the mean-shift method to seek on overall confidence map and puts the letter peak value, this position both be considered to the position pos=of target in present frame (x, y)=meanshift (P);
Described weight step of updating is:
Step f, sampling obtains positive and negative feedback data to testing result, calculates classification thresholds T k
Step g is used T kAlign the negative feedback value and carry out the resolution capability of partition test grader, and calculate the confidence level of grader in view of the above;
Step h upgrades the weight of each piece, resets the grader update mark.
2. the grader step of updating of the real-time tracking method of nonspecific target based on piecemeal according to claim 1, it is characterized in that, among the described step a, automatically the target area is divided into the piece of even big or small non-overlapping copies according to given target sizes, they do not need to have semantic feature, cut apart very simply arbitrary target all is suitable for, each other related of piece and piece also only is the space syntople, and each piece center and target's center have a fixing side-play amount
Figure FSA00000130688200011
3. the target detection step of the real-time tracking method of nonspecific target based on piecemeal according to claim 1 is characterized in that, among described step c, the d, with the shot chart I of the detection output of grader kBe converted to confidence map, this confidence map can better be described the probability of target in certain position:
P ( i , j ) = Σ k = 1 K ω k p k ( i + x k , j + y k ) Wherein p k ( i , j ) = 1 1 + exp ( - 2 I k ( i , j ) ) .
4. the real-time tracking method of nonspecific target based on piecemeal according to claim 1, it is characterized in that, described step f, g, h, after the target detection step is determined the target location, according to the resolution capability of testing result assessment grader, upgrade the weight of each piece then in view of the above again at present frame.
5. the weight step of updating of the real-time tracking method of nonspecific target based on piecemeal according to claim 1 is characterized in that, among the described step f, with pos+ (x k, y k) near r 1In the individual pixel coverage is the positive feedback value, r 2What individual pixel coverage was outer carries out grader C for the negative feedback value kResolving power check data collection.
6. the weight step of updating of the real-time tracking method of nonspecific target based on piecemeal according to claim 1 is characterized in that, among the described step f, with Gaussian Profile match positive feedback value and negative feedback value, utilizes its distributed constant (μ then 1, σ 1μ 0, σ 0) classification thresholds of calculating
Figure FSA00000130688200021
7. the weight step of updating of the real-time tracking method of nonspecific target based on piecemeal according to claim 1 is characterized in that, among described step g, the h, with the log-likelihood function of grader at the classification performance of present frame Weigh grader C kConfidence level, weight is upgraded rule and is:
Figure FSA00000130688200023
Hysteresis factors α is usually near 1, and is steady to guarantee that weight changes.
8. the weight step of updating of the real-time tracking method of nonspecific target based on piecemeal according to claim 1 is characterized in that, among the described step h, a grader update mark is set, as grader C kWhen the error rate of present frame is too high, suspends grader and upgrade.
CN2010101848768A 2010-05-25 2010-05-25 Real-time tracking method of nonspecific target based on partitioning Expired - Fee Related CN101867699B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010101848768A CN101867699B (en) 2010-05-25 2010-05-25 Real-time tracking method of nonspecific target based on partitioning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010101848768A CN101867699B (en) 2010-05-25 2010-05-25 Real-time tracking method of nonspecific target based on partitioning

Publications (2)

Publication Number Publication Date
CN101867699A true CN101867699A (en) 2010-10-20
CN101867699B CN101867699B (en) 2012-05-30

Family

ID=42959245

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010101848768A Expired - Fee Related CN101867699B (en) 2010-05-25 2010-05-25 Real-time tracking method of nonspecific target based on partitioning

Country Status (1)

Country Link
CN (1) CN101867699B (en)

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779267A (en) * 2011-05-12 2012-11-14 株式会社理光 Method and device for detection of specific object region in image
CN104484889A (en) * 2014-12-15 2015-04-01 三峡大学 Target tracking method and device
CN104504395A (en) * 2014-12-16 2015-04-08 广州中国科学院先进技术研究所 Method and system for achieving classification of pedestrians and vehicles based on neural network
CN104537387A (en) * 2014-12-16 2015-04-22 广州中国科学院先进技术研究所 Method and system for classifying automobile types based on neural network
CN104718559A (en) * 2012-10-22 2015-06-17 诺基亚技术有限公司 Classifying image samples
CN104820996A (en) * 2015-05-11 2015-08-05 河海大学常州校区 Target tracking method based on self-adaptive blocks of video
CN104992453A (en) * 2015-07-14 2015-10-21 国家电网公司 Target tracking method under complicated background based on extreme learning machine
CN105225253A (en) * 2015-09-22 2016-01-06 济南大学 A kind of method for tracking target based on Bayesian frame and Fast Fourier Transform (FFT)
CN105809136A (en) * 2016-03-14 2016-07-27 中磊电子（苏州）有限公司 Image data processing method and image data processing system
CN105898134A (en) * 2015-11-15 2016-08-24 乐视移动智能信息技术（北京）有限公司 Image acquisition method and device
CN106295548A (en) * 2016-08-05 2017-01-04 西北工业大学 A kind of method for tracking target and device
CN107766823A (en) * 2017-10-25 2018-03-06 中国科学技术大学 Anomaly detection method in video based on key area feature learning
CN108416376A (en) * 2018-02-27 2018-08-17 北京东方天得科技有限公司 It is a kind of that system and method is managed in way logistics people's vehicle tracing and monitoring based on SVM
CN108447045A (en) * 2018-02-02 2018-08-24 北京理工雷科电子信息技术有限公司 A kind of SAR remote sensing images waters detection method based on SAT integrograms
CN109272016A (en) * 2018-08-08 2019-01-25 广州视源电子科技股份有限公司 Object detection method, device, terminal device and computer readable storage medium
CN109325519A (en) * 2018-08-20 2019-02-12 中国铁道科学研究院集团有限公司电子计算技术研究所 Fault recognition method and device
CN110033473A (en) * 2019-04-15 2019-07-19 西安电子科技大学 Motion target tracking method based on template matching and depth sorting network
CN111738218A (en) * 2020-07-27 2020-10-02 成都睿沿科技有限公司 Human body abnormal behavior recognition system and method
CN112418120A (en) * 2020-11-27 2021-02-26 湖南师范大学 Crowd detection method based on peak confidence map
CN112862863A (en) * 2021-03-04 2021-05-28 广东工业大学 Target tracking and positioning method based on state machine

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6526161B1 (en) * 1999-08-30 2003-02-25 Koninklijke Philips Electronics N.V. System and method for biometrics-based facial feature extraction
CN101142593A (en) * 2005-03-17 2008-03-12 英国电讯有限公司 Method of tracking objects in a video sequence
CN101281648A (en) * 2008-04-29 2008-10-08 上海交通大学 Method for tracking dimension self-adaption video target with low complex degree
WO2009082700A1 (en) * 2007-12-21 2009-07-02 Honda Motor Co., Ltd. Online articulate object tracking with appearance and shape

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6526161B1 (en) * 1999-08-30 2003-02-25 Koninklijke Philips Electronics N.V. System and method for biometrics-based facial feature extraction
CN101142593A (en) * 2005-03-17 2008-03-12 英国电讯有限公司 Method of tracking objects in a video sequence
WO2009082700A1 (en) * 2007-12-21 2009-07-02 Honda Motor Co., Ltd. Online articulate object tracking with appearance and shape
CN101281648A (en) * 2008-04-29 2008-10-08 上海交通大学 Method for tracking dimension self-adaption video target with low complex degree

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
《BRITISH TELECOM (BT) TECHNOLOGY JOURNAL》 20040731 XU L-Q, LANDABASO J L, LEI B Segmentation and tracking of multiple moving objects for intelligent video analysis 全文 1-8 第22卷, 第3期 2 *
《PROC 3RD IEEE INTL WORKSHOP ONPERFORMANCE EVALUATION OF TRACKING AND SURVEILLANCE (PETS2002), COPENHAGEN, DENMARK》 20020630 SENIOR ANDREWS Tracking people with probabilistic appearance models 全文 1-8 , 2 *

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779267A (en) * 2011-05-12 2012-11-14 株式会社理光 Method and device for detection of specific object region in image
CN104718559A (en) * 2012-10-22 2015-06-17 诺基亚技术有限公司 Classifying image samples
US10096127B2 (en) 2012-10-22 2018-10-09 Nokia Technologies Oy Classifying image samples
CN104484889A (en) * 2014-12-15 2015-04-01 三峡大学 Target tracking method and device
CN104504395A (en) * 2014-12-16 2015-04-08 广州中国科学院先进技术研究所 Method and system for achieving classification of pedestrians and vehicles based on neural network
CN104537387A (en) * 2014-12-16 2015-04-22 广州中国科学院先进技术研究所 Method and system for classifying automobile types based on neural network
CN104820996B (en) * 2015-05-11 2018-04-03 河海大学常州校区 A kind of method for tracking target of the adaptive piecemeal based on video
CN104820996A (en) * 2015-05-11 2015-08-05 河海大学常州校区 Target tracking method based on self-adaptive blocks of video
CN104992453B (en) * 2015-07-14 2018-10-23 国家电网公司 Target in complex environment tracking based on extreme learning machine
CN104992453A (en) * 2015-07-14 2015-10-21 国家电网公司 Target tracking method under complicated background based on extreme learning machine
CN105225253A (en) * 2015-09-22 2016-01-06 济南大学 A kind of method for tracking target based on Bayesian frame and Fast Fourier Transform (FFT)
CN105898134A (en) * 2015-11-15 2016-08-24 乐视移动智能信息技术（北京）有限公司 Image acquisition method and device
CN105809136A (en) * 2016-03-14 2016-07-27 中磊电子（苏州）有限公司 Image data processing method and image data processing system
US10692217B2 (en) 2016-03-14 2020-06-23 Sercomm Corporation Image processing method and image processing system
CN106295548A (en) * 2016-08-05 2017-01-04 西北工业大学 A kind of method for tracking target and device
CN107766823A (en) * 2017-10-25 2018-03-06 中国科学技术大学 Anomaly detection method in video based on key area feature learning
CN108447045A (en) * 2018-02-02 2018-08-24 北京理工雷科电子信息技术有限公司 A kind of SAR remote sensing images waters detection method based on SAT integrograms
CN108447045B (en) * 2018-02-02 2022-03-25 北京理工雷科电子信息技术有限公司 SAR remote sensing image water area detection method based on SAT integral graph
CN108416376A (en) * 2018-02-27 2018-08-17 北京东方天得科技有限公司 It is a kind of that system and method is managed in way logistics people's vehicle tracing and monitoring based on SVM
CN108416376B (en) * 2018-02-27 2021-03-12 北京东方天得科技有限公司 SVM-based logistics man-vehicle tracking monitoring management system and method
CN109272016A (en) * 2018-08-08 2019-01-25 广州视源电子科技股份有限公司 Object detection method, device, terminal device and computer readable storage medium
CN109325519A (en) * 2018-08-20 2019-02-12 中国铁道科学研究院集团有限公司电子计算技术研究所 Fault recognition method and device
CN110033473A (en) * 2019-04-15 2019-07-19 西安电子科技大学 Motion target tracking method based on template matching and depth sorting network
CN110033473B (en) * 2019-04-15 2021-04-20 西安电子科技大学 Moving target tracking method based on template matching and depth classification network
CN111738218A (en) * 2020-07-27 2020-10-02 成都睿沿科技有限公司 Human body abnormal behavior recognition system and method
CN112418120A (en) * 2020-11-27 2021-02-26 湖南师范大学 Crowd detection method based on peak confidence map
CN112418120B (en) * 2020-11-27 2021-09-28 湖南师范大学 Crowd detection method based on peak confidence map
CN112862863A (en) * 2021-03-04 2021-05-28 广东工业大学 Target tracking and positioning method based on state machine
CN112862863B (en) * 2021-03-04 2023-01-31 广东工业大学 Target tracking and positioning method based on state machine

Also Published As

Publication number Publication date
CN101867699B (en) 2012-05-30

Similar Documents

Publication Publication Date Title
CN101867699B (en) Real-time tracking method of nonspecific target based on partitioning
CN109190544B (en) Human identity recognition method based on sequence depth image
CN103226835B (en) Based on method for tracking target and the system of online initialization gradient enhancement regression tree
CN106570490B (en) A kind of pedestrian's method for real time tracking based on quick clustering
CN110232308B (en) Robot-following gesture track recognition method based on hand speed and track distribution
CN102496016B (en) Infrared target detection method based on space-time cooperation framework
CN103793926B (en) Method for tracking target based on sample reselection procedure
CN103268616A (en) Multi-feature multi-sensor method for mobile robot to track moving body
CN103426179B (en) A kind of method for tracking target based on mean shift multiple features fusion and device
CN107798691B (en) A kind of unmanned plane independent landing terrestrial reference real-time detection tracking of view-based access control model
CN103886325A (en) Cyclic matrix video tracking method with partition
CN104616319A (en) Multi-feature selection target tracking method based on support vector machine
CN110991397B (en) Travel direction determining method and related equipment
CN103593679A (en) Visual human-hand tracking method based on online machine learning
CN103105924A (en) Man-machine interaction method and device
CN104463909A (en) Visual target tracking method based on credibility combination map model
CN105809713A (en) Object tracing method based on online Fisher discrimination mechanism to enhance characteristic selection
Xiao et al. Traffic sign detection based on histograms of oriented gradients and boolean convolutional neural networks
Qing et al. A novel particle filter implementation for a multiple-vehicle detection and tracking system using tail light segmentation
Adiwinata et al. Fish species recognition with faster r-cnn inception-v2 using qut fish dataset
CN103092925B (en) A kind of video abstraction generating method and device
CN103996207A (en) Object tracking method
CN102663773A (en) Dual-core type adaptive fusion tracking method of video object
CN106485283B (en) A kind of particle filter pedestrian target tracking based on Online Boosting
Qin et al. Multi-scaling detection of singular points based on fully convolutional networks in fingerprint images

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

Granted publication date: 20120530

Termination date: 20150525

EXPY Termination of patent right or utility model