CN107358620A - Pedestrian tracting method and its device are blocked based on hybrid system entirely - Google Patents

Pedestrian tracting method and its device are blocked based on hybrid system entirely Download PDF

Info

Publication number
CN107358620A
CN107358620A CN201610303631.XA CN201610303631A CN107358620A CN 107358620 A CN107358620 A CN 107358620A CN 201610303631 A CN201610303631 A CN 201610303631A CN 107358620 A CN107358620 A CN 107358620A
Authority
CN
China
Prior art keywords
pedestrian
model
tracking
identification
feature
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
CN201610303631.XA
Other languages
Chinese (zh)
Other versions
CN107358620B (en
Inventor
胡士强
张晓宇
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.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
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 Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN201610303631.XA priority Critical patent/CN107358620B/en
Publication of CN107358620A publication Critical patent/CN107358620A/en
Application granted granted Critical
Publication of CN107358620B publication Critical patent/CN107358620B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)

Abstract

It is a kind of that pedestrian tracting method is blocked based on hybrid system entirely, need according to pedestrian movement's feature and in the track the pedestrian's feature learnt, build corresponding trace model and identification model, when pedestrian is not blocked, learn pedestrian's feature while trace model tracking pedestrians, perceiving pedestrian by the sensed condition of Model transfer rule when pedestrian is blocked is blocked and switches to identification model；Identification model constantly recognizes mistake with pedestrian from testing result using the pedestrian's feature learnt according to the identification condition of Model transfer rule；When identification model recognizes mistake with pedestrian, by the states of the replacements Conditional reset tracking pedestrians of Model transfer rule and it is switched to trace model and continues to track, realization has the pedestrian tracking under full circumstance of occlusion；The present invention is reasonable in design, and model-driven and data-driven algorithm are combined using the thought of hybrid system and solve this practical problem, substantially increases the degree of accuracy and the success rate of tracking.

Description

Pedestrian tracting method and its device are blocked based on hybrid system entirely
Technical field
It is specifically a kind of based on hybrid system the present invention relates to a kind of technology in computer vision monitoring objective tracking field Block pedestrian tracting method and its device entirely.
Background technology
Traditional single pedestrian tracking mainly has two major class methods：Object representation and positioning, filtering and data correlation.It is conventional at present Pedestrian tracking algorithm be first kind method.This kind of method can be divided into three group methods again：Production tracking, discriminate tracking and inspection Survey with tracking hybrid tracking.The representative model of production Algorithm Learning one is represented tracking target and caused using the pattern search The minimum image-region of reconstruction error；Tracking problem is considered as two classification problems in local search area by discriminate algorithm, profit With the grader learnt the image-region with background most discrimination is obtained in current frame search.
Tremendous development has been obtained in last decade based on object representation and the algorithm of positioning.Proposed for blocking for pedestrian Effective algorithm, such as L1 track algorithms (Xue Mei and Haibin Ling, Robust Visual Tracking and Vehicle Classification via Sparse Representation[J].IEEE TPAMI,November 2011,vol.33(11):2259-2271) Tracking is converted into the sparse Approximation Problem under particle filter framework.During tracking, candidate target utilizes To Template and trifling Template carries out linear expression.Block occur when, limited sparse template will be activated but whole coefficient vector remain in that it is sparse. FCT(Kaihua Zhang,Lei Zhang and Ming-Hsuan Yang,Fast compressive tracking[J].IEEE T PAMI, October 2014,vol.36(10):2002-2015) model is the apparent mould based on the non-self-adapting Random Maps for retaining picture structure Type.One very sparse measurement matrix extracts effective compressive features from prospect and target context, and the task of tracking is by a pressure Two graders with online updating function are completed in contracting domain.Apparent model from incoherent measurement learning to target and background Otherness, the influence of background pixel can be reduced, so as to handle the problem of blocking and tracking target carriage change.For total It, both approaches using them there is the apparent expression of target of distinction and the effective process part of model modification mechanism to block even sternly Block again.Block occur when, they take full advantage of the part that is not blocked target signature tracking target.But when target is complete After blocking, if the apparent model of target stops template renewal, then calculated when losing target and reappearing in the range of target search Method can continue to track target.If deviation occurs in template renewal or loss target is appeared in beyond target search scope, algorithm will Failure.
The representative algorithm that detect and track combines is TLD (Zdenek Kalal, Krystian Mikolajczyk, Jiri Matas. Tracking-learning-detection[J].IEEE T PAMI,July 2012,34(7):1409-1422) algorithm.Detected in the algorithm Device and tracker estimate the position candidate of target simultaneously.The output tracking result when tracking result is more effective, when testing result is more effective When reset target position.Learner utilizes the more preferable detector of best estimate result repetitive exercise.The algorithm has very to rigidity target Good ability of tracking, but it is bad that the pedestrian tracking effect blocked entirely is particularly present for pedestrian tracking.Filtering and the side of data correlation The method commonly used in method for monotrack has PF filtering and Kalman filter and its derivative EKF filtering and UKF filtering. These algorithm combination gate techniques can realize the reliable judgement that target is lost.But when target there is a situation where to block entirely, due to mesh The loss of mark movable information causes follow-up tracking failure.In order to realize that pedestrian blocks front and rear continuous tracking, it is necessary to 1) be hidden in pedestrian Good tracking pedestrians and some are trained to recognize the feature of the pedestrian so that the pedestrian can be had when occurring again after blocking before gear Effect identification；2) tracking based on pedestrian's state and the study to pedestrian's feature are stopped after pedestrian is blocked；3) spy learnt is utilized Sign recognizes mistake from testing result and is tracked with resetting tracking mode during pedestrian.
Hybrid system is that one kind had not only included continuous dynamic behaviour but also the system for including Discrete Dynamic behavior, these continuous and discrete powers Scholarship and moral conduct interacts not only to exist jointly.The development and evolution of hybrid system had both depended on the response to discrete temporal event, The response of the dynamic behavior changed over time represented differential and difference equation is depended on again.This method is widely used in calculating Machine scientific domain, modeling and simulation field, system and control scientific domain (Arjian van der Schaft and Hans Schumacher. An introduction to hybrid dynamical systems [B] .1st ed.vol.251.Springer, London, UK, 2000.), But the algorithm is seldom used in video frequency object tracking field.
Due to single camera visual angle limitation in monitoring scene, the full circumstance of occlusion of pedestrian inevitably be present, in this case Just need algorithm to recognize to pedestrian tracking to block and continue to track when losing and again occurring with pedestrian.
Found by the retrieval to prior art, Chinese patent literature CN102663409A, date of publication 2012.9.12, it is public A kind of pedestrian tracting method based on HOG-LBP descriptions has been opened, has been comprised the following steps：A1, Sample Establishing；A2, feature extraction； A3, establish SVM models；A4, classifier training；A5, video capture and pretreatment；A6, video pedestrian detection；A7, regard Frequency pedestrian tracking：The step A6 pedestrians detected are tracked using based on the particle filter tracking method of HOG-LBP features. But the technology 1) due to the defects of HOG-LBP description add the pedestrian detection algorithm of SVM classifier, it is past when choosing positive negative sample Toward the training sample for needing addition tracking pedestrians, it is unfavorable for the detection to unknown pedestrian；2) do not account for going when solving occlusion issue People falls the situation outside particle filter estimation range；3) algorithm does not have flight path termination mechanism, still updates mould after pedestrian is blocked Plate, template renewal mistake is inevitably resulted in, follow-up pedestrian tracking can not be carried out；4) if one is blocked entirely according to arthmetic statement pedestrian The position that particle filter does not predict is appeared in after the section time, is easily identified as newborn target.
The content of the invention
The present invention be directed to deficiencies of the prior art, propose it is a kind of based on hybrid system block full pedestrian tracting method and Its device, by building pedestrian tracking model and pedestrian's identification model, before pedestrian is by complete block by pedestrian tracking model following simultaneously Learn pedestrian's feature, when tracking pedestrians are blocked entirely stop study, by pedestrian's identification model using the pedestrian's feature learnt from Mistake is picked out in testing result with pedestrian and continues to track.
The present invention is achieved by the following technical solutions：
Pedestrian tracting method is blocked based on hybrid system entirely the present invention relates to a kind of, according to pedestrian movement's feature and in the track The pedestrian's feature learnt is needed, builds corresponding trace model and identification model, when pedestrian is not blocked, trace model tracking pedestrians While learn pedestrian's feature, perceiving pedestrian by the sensed condition of Model transfer rule when pedestrian is blocked is blocked and switches to and distinguish Know model；Identification model using the pedestrian's feature learnt, constantly lose with pedestrian from testing result by identification；Recognized in identification model When losing with pedestrian, according to the state of the replacement Conditional reset tracking pedestrians of Model transfer rule and it is switched to trace model and continues to track, Realize the pedestrian tracking existed under full circumstance of occlusion.
Described testing result refers to：Pedestrian area is detected using pedestrian detection algorithm ACF.
Described pedestrian detection result is expressed asWherein：For the position of noisy i-th of the measurement of t band And yardstick.
Described trace model is expressed asWherein：X is pedestrian movement's state parameter,Needs are trained Specific pedestrian's grader,For pedestrian's average movement velocity,For the average dimension information of tracking pedestrians.
Described identification model includes：Reachable sub-model, yardstick submodel and the apparent submodel of pedestrian movement.
Described reachable sub-model isWherein：diFor testing result(1:2) center with Tracking pedestrians center x before the failure of track modelk(1:2) Euclidean distance, ifThen set up up to sub-model.
Described yardstick submodel is P 't|t-1=Q '+F ' P 't-1FT, S '=H ' P 't|t-1H′T+ R ', wherein：Z ' is containing noisy On the dimensional information of testing result, F '=I2, H '=I2,If γ ' is two-dimentional ellipsoid door thresholding, then yardstick submodel is set up.
Described apparent submodel exists Serve as reasonsThe pedestrian image area confined Set up during domain.
The present invention relates to a kind of device for realizing the above method, including：Video acquisition unit, tracking and feature learning unit, distinguish Know unit and transition rule unit, wherein：Tracking and feature learning unit learn pedestrian's feature while tracking pedestrians, and identification is single Member according to tracking and feature learning modular learning to pedestrian's feature recognize from testing result lose with pedestrian；Transition rule unit according to Sensed condition, identification condition and replacement condition are tracked switching and state replacement between feature learning unit and identification unit.
Technique effect
Compared with prior art, the present invention is reasonable in design, and identification is tracked to the pedestrian blocked entirely be present based on hybrid system, It is easily controllable, the very big degree of accuracy and the success rate for improving pedestrian tracking, and this algorithm does not need the sample information of tracking pedestrians It is trained, is the detection that unknown pedestrian can be achieved using advanced pedestrian detection algorithm ACF combination INRIA databases.
Brief description of the drawings
Fig. 1 is to block pedestrian tracting method schematic diagram entirely；
Fig. 2 is embodiment qualitative results figure；
Fig. 3 is to block pedestrian tracking schematic device entirely.
Embodiment
Embodiments of the invention are elaborated below, the present embodiment is implemented under premised on technical solution of the present invention, Detailed embodiment and specific operating process are given, but protection scope of the present invention is not limited to following embodiments.
Embodiment 1
As shown in figure 1, the present embodiment, which is related to a kind of complete based on hybrid system, blocks pedestrian tracting method, comprise the following steps：
Step 1, according to the pedestrian area of the pedestrian detection algorithm ACF N number of pedestrians detected as testing result, based on pedestrian Motion conform generally to constant speed motion model it is assumed that with reference to Kalman filter algorithm build trace model, realize pedestrian tracking, and Learn pedestrian's feature.
Described pedestrian detection algorithm ACF (Aggregated Channel Features) is quoted from Piotr Dollar's et al.《Fast feature pyramids for object detection》([J].IEEET PAMI,August 2014,36(8):A 1532-1545) text.
Described pedestrian detection result is expressed asWherein：Represent the position of noisy i-th of the measurement of t band Put and yardstick.
Described pedestrian tracking modelState at a timeWherein：X is pedestrian movement's state Parameter,To need the pedestrian's grader trained,For pedestrian's average movement velocity,For the average dimension of tracking pedestrians.
Described pedestrian movement's state parameterWherein：(pw,ph) represent pedestrian position,The speed of pedestrian is represented, (w, h) represents that the width of pedestrian is high.
Described pedestrian movement's state parameter x motion model and measurement model be respectively： Wherein：Represent that average and variance are respectively m and P Gaussian density function, F For state-transition matrix, Q is process noise covariance matrix, and H is observing matrix, and R is process noise covariance.
Because the motion of pedestrian often meets constant speed motion, thus：
Wherein：σuFor process noise standard deviation, σvFor observation noise standard deviation.
Described pedestrian's graderSpecifically refer to：While pedestrian tracking, N is extracted near tracking resultposIndividual positive sample And NnegIndividual negative sample is simultaneously worth accordingly according to formwork calculation, and trains one using increment SVM based on these values and be based on pedestrian's feature The grader that tracking pedestrians can be identified from testing result.
Centered on estimating pedestrian center, the border circular areas using the Sw of tracking pedestrians width as radius extracts described positive sample, Negative sample is centered on estimating pedestrian center, and the circle ring area using the Swi of tracking pedestrians width and Swo as interior outer radius is taken out respectively Take.
Described positive sample and the value of negative sample represent with the positive negative sample and the similarity sim (T, y) of template extracted, wherein：T For template image, y is the image of positive sample and negative sample.
The image y of described template image T and positive sample and negative sample (such as 128 × 64) after identical image size is adjusted to, Pedestrian's measuring similarity is carried out using pedestrian's measuring similarity algorithm or similar means.
Described pedestrian's measuring similarity algorithm is quoted from Zhao's et al.《Unsupervised Salience Learning for Person Re-identification》(In proceeding of Conferenceon Computer Vision and Pattern Recognition[C],Portland,OR,USA,25-27June 2013:A 3586-3593) text.
Described increment type SVM can use Diehl's et al.《SVM incremental learning,adaptation and optimization》(In Proceedings of the International Joint Conference on Neural Networks[C], Portland,OR,USA,20-24July 2003:2685-2690)。
Described pedestrian's average movement velocityWherein：V is the current movement velocity of pedestrian, and α is pedestrian Average speed learning parameter, 0≤α≤1.
The average dimension of described tracking pedestriansWherein：S is the current yardstick of pedestrian, and β is pedestrian Average dimension learning parameter, 0≤β≤1.
Step 2, structure identification model, are sieved with reference to the accessibility of pedestrian movement, dimensional information and apparent information from testing result Select the target possibly as tracking pedestrians.
Described identification modelIncluding：Reachable sub-model, yardstick submodel and the apparent submodel of pedestrian movement.
Described reachable sub-model isWherein：diFor testing resultCenter and row Tracking pedestrians center x before the failure of people's trace modelk(1:2) Euclidean distance, ifThen up to sub-model into It is vertical.
Described yardstick submodel is P 't|t-1=Q '+F ' P 't-1FT, S '=H ' P 't|t-1H′T+ R ', wherein：Z ' is containing noisy On the dimensional information of testing result, F '=I2, H '=I2,If γ ' is two-dimentional ellipsoid door thresholding, then yardstick submodel is set up.
Described yardstick submodel can use ellipsoid door algorithm, quoted from Blackman's et al.《Design and analysis of modern tracking sysyems》(Artech House, Boston, 1999) one text.
Described two-dimentional ellipsoid door thresholding is quoted from Zhou Hongren's et al.《Maneuvering target tracking》([B] National Defense Industry Press, 1991) One book.
Described apparent submodel existsShi Chengli.WhereinFor the i-th of t Individual pedestrian byThe image information confined,For the specific pedestrian's grader arrived in trace model learning.
Step 3, the Model transfer rule model that simultaneously real-time judge current time works is created, and carry out experiment detection.
Described Model transfer rule includes sensed condition τ1, identification condition τ2With replacement condition τ3。
Described sensed condition τ1For：IfAnd the testing result of existence anduniquess meets Pt|t-1=Q+FPt-1FT,(γ is four-dimensional ellipsoid door thresholding), thenXt=Xt-1。
Described identification condition τ2For：IfIf pedestrian's identification modelDo not set up, thenXt=Xt-1。
Described mould resets condition τ3For：IfIf pedestrian's identification modelSet up, then
The video that the data of described experiment detection are blocked entirely using the presence pedestrian of self-timer.
Described video posture before and after tracking pedestrians enter group and go out group varies widely.
For described video after tracking pedestrians are by complete block, interference pedestrian first goes out and lost the Location-Unknown for going out group with pedestrian.
The parameter value of the present embodiment is as shown in table 1.
The parameter setting table of table 1
As shown in Fig. 2 being the qualitative experiment result of the present embodiment, the present embodiment can ensure that tracking pedestrians lose front as seen from the figure Stable tracking；It can judge to block in time after tracking pedestrians are blocked；Do not tracked when disturbing pedestrian to occur, in real mesh Mistake is recognized in time with pedestrian after marking now and is tracked；Is there is not with pedestrian position and is influenceed that (pedestrian is hidden by losing again in the algorithm Still by backtracking, algorithm can recognize and track again mistake with pedestrian for direct of travel walking either along before blocking after gear).
The present embodiment and mean center application condition of FCT, LIAPG and TLD algorithm in the five groups of test videos such as institute of table 2 Show.
The mean center application condition table of table 2
Test1 Test2 Test3 Test4
FCT 28 40 64 17
TLD 45 49 26 13
L1APG 20 44 41 7
Ours 2 3 2 4
As shown in Table 2, due to can accurately capture mistake after blocking completely from testing result with pedestrian, the present embodiment is every Mean center error in group test is minimum.Other algorithms are in average due to hunting zone or to pedestrian's identification capability difference Heart error is higher.
Tracking success rate result of the present embodiment with FCT, LIAPG and TLD algorithm in four groups of test videos is as shown in table 3.
Table 3 tracks success rate result
Test1 Test2 Test3 Test4
FCT 0.63 0.60 0.01 0.57
TLD 0.36 0.52 0.70 0.74
L1APG 0.65 0.60 0.53 0.91
Ours 0.99 0.93 0.96 0.93
It can be obtained by table 3, the present embodiment pedestrian tracking success rate highest when complete block be present, and keep stable in testing, It can realize that tracking pedestrians timely track after disappearing to terminate, and accurate recognition with target and can be carried out to mistake from testing result Tracking.
The Model transfer rule of the present embodiment uses ellipsoid door algorithm, when testing result not falls in the ellipse of tracking pedestrians for only one When in goal, model needs to be transferred to identification model from trace model；Recognized in identification model fails from its testing result mistake with Model jumps to identification model again during pedestrian；When identification model recognizes tracking pedestrians from its testing result, reset pedestrian with Track location status is simultaneously transferred to trace model from identification model.
As shown in figure 3, the present embodiment is related to a kind of device for realizing the above method, including：Video acquisition unit, tracking and spy Unit, identification unit and transition rule unit are levied, wherein：Tracking cell learns pedestrian's feature while tracking pedestrians, distinguishes Know unit according to tracking and feature learning modular learning to pedestrian's feature recognize to testing result lose with pedestrian；Transition rule list Member is tracked switching and state weight between feature learning unit and identification unit according to sensed condition, identification condition and replacement condition Put.

Claims (8)

1. a kind of block pedestrian tracting method entirely based on hybrid system, it is characterised in that according to pedestrian movement's feature and is tracking The middle pedestrian's feature for needing to learn, builds corresponding trace model and identification model, when pedestrian is not blocked, trace model tracking lines Learn pedestrian's feature while people, be blocked and switch to by the sensed condition perception pedestrian of Model transfer rule when pedestrian is blocked Identification model；Identification model using the pedestrian's feature learnt, constantly lose with pedestrian from testing result by identification；Recognized in identification model To when losing with pedestrian, by the states of the replacement Conditional reset tracking pedestrians of Model transfer rule and it is switched to trace model and continues to track, Realize the pedestrian tracking existed under full circumstance of occlusion；
Described testing result refers to：The pedestrian area detected using pedestrian detection algorithm ACF.
2. according to claim 1 block pedestrian tracting method entirely, it is characterized in that, the testing result of described pedestrian is expressed asWherein：Position and yardstick for noisy i-th of the measurement of t band.
3. according to claim 1 block pedestrian tracting method entirely, it is characterized in that, described trace model is expressed asWherein：X is pedestrian movement's state parameter,To need the pedestrian's grader trained,Averagely transported for pedestrian Dynamic speed,For the average dimension information of tracking pedestrians.
4. according to claim 1 block pedestrian tracting method entirely, it is characterized in that, described identification model includes：Pedestrian transports Dynamic reachable sub-model, yardstick submodel and apparent submodel.
5. according to claim 1 block pedestrian tracting method entirely, it is characterized in that, described reachable sub-model isWherein：diFor testing resultBefore center is failed with pedestrian tracking model in tracking pedestrians Heart xk(1:2) Euclidean distance, ifThen set up up to sub-model.
6. according to claim 1 block pedestrian tracting method entirely, it is characterized in that, described yardstick submodel is P′t|t-1=Q '+F ' P 't-1FT, S '=H ' P 't|t-1H′T+ R ', wherein：Z ' be containing the noisy dimensional information on testing result, F '=I2, H '=I2,Ifγ ' is two-dimentional ellipsoid door thresholding, Then yardstick submodel is set up.
7. according to claim 1 block pedestrian tracting method entirely, it is characterized in that, described apparent submodel existsShi Chengli.WhereinServe as reasonsThe pedestrian image region confined,For in tracking mould Specific pedestrian's grader that type learning arrives.
8. a kind of device for realizing the above method, including：Video acquisition unit, tracking and feature learning unit, identification unit and turn Rules unit is moved, wherein：Tracking and feature learning unit learn pedestrian's feature while pedestrian tracking, and identification unit is according to feature Pedestrian's feature that unit learns establishes identification model and identification in testing result is lost with pedestrian；Transition rule unit is according to sense Know condition, identification condition and reset switching and state replacement that condition is tracked between unit and identification unit.
CN201610303631.XA 2016-05-10 2016-05-10 Hybrid system-based full-coverage pedestrian tracking method and device Active CN107358620B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610303631.XA CN107358620B (en) 2016-05-10 2016-05-10 Hybrid system-based full-coverage pedestrian tracking method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610303631.XA CN107358620B (en) 2016-05-10 2016-05-10 Hybrid system-based full-coverage pedestrian tracking method and device

Publications (2)

Publication Number Publication Date
CN107358620A true CN107358620A (en) 2017-11-17
CN107358620B CN107358620B (en) 2020-07-10

Family

ID=60271264

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610303631.XA Active CN107358620B (en) 2016-05-10 2016-05-10 Hybrid system-based full-coverage pedestrian tracking method and device

Country Status (1)

Country Link
CN (1) CN107358620B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007033286A2 (en) * 2005-09-13 2007-03-22 Verificon Corporation System and method for object tracking and activity analysis
CN102646279A (en) * 2012-02-29 2012-08-22 北京航空航天大学 Anti-shielding tracking method based on moving prediction and multi-sub-block template matching combination
CN102881022A (en) * 2012-07-20 2013-01-16 西安电子科技大学 Concealed-target tracking method based on on-line learning
CN104951758A (en) * 2015-06-11 2015-09-30 大连理工大学 Vehicle-mounted method and vehicle-mounted system for detecting and tracking pedestrians based on vision under urban environment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007033286A2 (en) * 2005-09-13 2007-03-22 Verificon Corporation System and method for object tracking and activity analysis
CN102646279A (en) * 2012-02-29 2012-08-22 北京航空航天大学 Anti-shielding tracking method based on moving prediction and multi-sub-block template matching combination
CN102881022A (en) * 2012-07-20 2013-01-16 西安电子科技大学 Concealed-target tracking method based on on-line learning
CN104951758A (en) * 2015-06-11 2015-09-30 大连理工大学 Vehicle-mounted method and vehicle-mounted system for detecting and tracking pedestrians based on vision under urban environment

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
AAYUSH ANKIT等: "A cascade framework for unoccluded and occluded pedestrian detection", 《PROCEEDINGS OF THE 2014 IEEE STUDENTS" TECHNOLOGY SYMPOSIUM》 *
苏松志等: "《行人检测 理论与实践》", 31 March 2016, 厦门大学出版社 *
陈宗海: "《***仿真技术及其应用 第7卷》", 31 August 2005, 中国科学技术大学出版社 *

Also Published As

Publication number Publication date
CN107358620B (en) 2020-07-10

Similar Documents

Publication Publication Date Title
CN111310583B (en) Vehicle abnormal behavior identification method based on improved long-term and short-term memory network
CN106408594B (en) Video multi-target tracking based on more Bernoulli Jacob's Eigen Covariances
CN106778712B (en) Multi-target detection and tracking method
CN105224912B (en) Video pedestrian's detect and track method based on movable information and Track association
CN103268616B (en) The moveable robot movement human body tracing method of multi-feature multi-sensor
CN110008867A (en) A kind of method for early warning based on personage's abnormal behaviour, device and storage medium
CN108537829B (en) Monitoring video personnel state identification method
CN103617636B (en) The automatic detecting and tracking method of video object based on movable information and sparse projection
CN107609597A (en) A kind of number of people in lift car detecting system and its detection method
CN103971386A (en) Method for foreground detection in dynamic background scenario
CN107491720A (en) A kind of model recognizing method based on modified convolutional neural networks
CN106127812B (en) A kind of passenger flow statistical method of the non-gate area in passenger station based on video monitoring
CN109684996A (en) Real-time vehicle based on video passes in and out recognition methods
Hu et al. Face Detection based on SSD and CamShift
CN108416258A (en) A kind of multi-human body tracking method based on human body model
CN104992453A (en) Target tracking method under complicated background based on extreme learning machine
Kim Multiple vehicle tracking and classification system with a convolutional neural network
CN106650805B (en) A kind of visual target tracking method and device
CN101408983A (en) Multi-object tracking method based on particle filtering and movable contour model
CN110728694A (en) Long-term visual target tracking method based on continuous learning
CN112541424A (en) Real-time detection method for pedestrian falling under complex environment
Yang et al. Vehicle detection methods from an unmanned aerial vehicle platform
CN111368770B (en) Gesture recognition method based on skeleton point detection and tracking
CN108764338A (en) A kind of pedestrian tracking algorithm applied to video analysis
CN104778699A (en) Adaptive object feature tracking method

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
GR01 Patent grant
GR01 Patent grant