CN105894535A - Bayes-based vortex automatic tracking method - Google Patents

Bayes-based vortex automatic tracking method Download PDF

Info

Publication number
CN105894535A
CN105894535A CN201610191428.8A CN201610191428A CN105894535A CN 105894535 A CN105894535 A CN 105894535A CN 201610191428 A CN201610191428 A CN 201610191428A CN 105894535 A CN105894535 A CN 105894535A
Authority
CN
China
Prior art keywords
vortex
tracker
observation
tracking
trail
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
CN201610191428.8A
Other languages
Chinese (zh)
Other versions
CN105894535B (en
Inventor
易嘉伟
杜云艳
周成虎
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.)
Institute of Geographic Sciences and Natural Resources of CAS
Original Assignee
Institute of Geographic Sciences and Natural Resources of CAS
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 Institute of Geographic Sciences and Natural Resources of CAS filed Critical Institute of Geographic Sciences and Natural Resources of CAS
Priority to CN201610191428.8A priority Critical patent/CN105894535B/en
Publication of CN105894535A publication Critical patent/CN105894535A/en
Application granted granted Critical
Publication of CN105894535B publication Critical patent/CN105894535B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10016Video; Image sequence

Abstract

The invention discloses a bayes-based vortex automatic tracking method. The method includes the steps of modeling a vortex moving process through Bayes-based kalman filtering, tracking and matching a predicted position and an observation result, and solving the matching conflict of multi-target vortex tracking through a Hungarian optimization algorithm. Matching conflict of a multi-target vortex tracking process is overcome, and the accuracy of a tracking result is improved.

Description

A kind of based on Bayesian vortex method for automatic tracking
Technical field
The present invention relates to a kind of based on Bayesian vortex method for automatic tracking, belong to ocean eddy automatic Extraction Algorithm field.
Background technology
Mesoscale eddy (abbreviation vortex) is widely distributed in ocean and very important a kind of dynamic phenomenon, it can not only transmit heat, Propagate nutrient substance, be also the major embodiment of ocean kinetic energy simultaneously.The development of Modern remote observation technology and application are research ocean The differentiation Evolution of vortex, it is provided that long-time abundant observational data.Excavate from the observational data of these magnanimity and extract The evolutionary process of vortex, is the basic premise of research eddying motion Evolution.
At present, the method for tracing of vortex evolutionary process mainly has three kinds: 1) range search method, mainly follow the trail of in vortex with current Heart point is reference settings search radius, falls into the vortex in radius in next moment search, the vortex that selected distance is nearest As developing follow-up (Chelton et al.2011);2) similarity follow the trail of, on the basis of range search method, add vortex it Between the tolerance of similarity degree, with develop about most like vortex follow-up (Chaigneau et al.2008);3) area overlap method, Before and after developing according to vortex, whether spatial shape exists overlap, judges whether to belong to a vortex evolutionary process (Henson et al.2008)。
These methods existing, algorithm idea is simple, is widely used, but there are 2 deficiencies: one have ignored vortex's motion No matter essence, be half path search or area overlap, if following the trail of the next moment with the position of current vortex for reference, this Assume that in matter that the vortex next moment remains in original place and do not moves, and this is not in accordance with facts;Two is when multiple vortex positions Close to time, easily form complicated tracking and matching problem, as two vortexs the next moment track three potential follow-up, study carefully Unexpectedly which kind of coupling just best suits vortex's motion feature, and this problem tackles way in existing three kinds of methods the most accordingly.
Summary of the invention
Present invention solves the technical problem that: provide a kind of based on Bayesian vortex method for automatic tracking, use Kalman filter Eddying motion process is modeled, the problem solving the prediction of vortex subsequent time position, uses Hungarian optimum simultaneously Change matching algorithm, solve matching conflict problem during multiple vortex target tracking, improve the accuracy following the trail of result.
Technical scheme: a kind of based on Bayesian vortex method for automatic tracking, first builds based on Bayesian The motor process of vortex is modeled by Kalman filter (Bar-Shalom et al.2001), predicts that vortex is next from probability The position of moment most probable appearance and effective error scope, then be tracked mating with the result of observation, chase after if there is multiple target Track mate, then use Hungarian optimization matching algorithm (J.Munkres 1957) search for each vortex target optimal after Continue.
Specifically comprise the following steps that
Step 1, initialization, set up an empty vortex tracker set, a vortex observation set and an empty tracking Results set.Each vortex tracker will record the follow-up pass of preamble of a vortex evolutionary process during automatic tracing System.Vortex observation set will record the eddy information (position, attribute etc.) that each moment observes.Tracking results set will The vortex tracker that storage record life process has been followed the trail of.The each vortex observing the first moment, builds vortex tracker, The positional information of record vortex, and create the Kalman filter of default parameters for tracking prediction, then tracker is added to In vortex tracker set;
Step 2, startup are followed the trail of, and from the next moment, temporally frame is circulated, if being recycled to last frame, chases after Track stops, and follows the tracks of the vortex tracker that results set is collected, and have recorded the Life Evolution procedural information following the trail of vortex;Otherwise, enter Row automatic tracing, performs step 3-4;
Step 3, all vortex observations of acquisition current time frame, add in observation set;
Step 4, data association, traversal tracker set in each vortex tracker, first carry out Kalman filter with Track is predicted, then by all vortexs in the vortex position of prediction and the vortex observation set of range of error and current time, carries out Association coupling:
If ● be not matched to the vortex in observation set in the range of the forecast error of vortex tracker, then by this tracker from Set is taken out, joins in the tracking results set that step 1 is set up, show that the Life Evolution process of this vortex has followed the trail of knot Bundle;
If ● have in the range of the forecast error of vortex tracker and only one coupling vortex observation, then from vortex observation Set is taken out this vortex, the follow-up vortex traced into as tracker, and use the Kalman filter observation to this vortex Position is modified, it is thus achieved that filtered vortex position and range of error, and the process trace after carrying out of taking this as the standard;
If ● there is the vortex observation of multiple coupling in the range of the forecast error of vortex tracker, then initially with Hungarian The bipartite graph (bipartite graph) that this tracker and vortex observation are constituted by algorithm carries out optimization matching, for matching The vortex tracker of observation, takes out this observation from observation set, the follow-up vortex traced into as tracker, and Use Kalman filter that the observation position of this vortex is modified, it is thus achieved that filtered vortex position and range of error, and Take this as the standard the process trace after carrying out;
● if observation set exists the vortex of non-matched jamming device, the most newly-built vortex tracker, add to vortex with In track device set.
Present invention advantage compared with prior art is: uses based on Bayesian Kalman filter, simulates whirlpool more accurately The motor process of rotation;Use Hungarian algorithm to solve a complex match conflict difficult problem for multi-target tracking, improve the standard of tracking Really property.
Accompanying drawing explanation
Fig. 1 is the flowchart of the present invention.
Detailed description of the invention
As it is shown in figure 1, the present invention to be embodied as step as follows:
Step 1, initialization, set up empty vortex tracker set (trackerList), a vortex observation set (observeList) and one empty tracking results set (completeList).Each vortex tracker is by the process at automatic tracing In by Ids attribute record vortex ID, storage vortex life process the follow-up relation of preamble.Vortex observation set will note Record the vortex position that each moment observes.Follow the tracks of the vortex tracker that storage record life process has been followed the trail of by results set. The each vortex observing the first moment, builds vortex tracker, records the position vector of vortex, and create in Obs attribute Build the Kalman filter of default parameters for tracking prediction, then tracker is added in vortex tracker set.
Tracker, observation data structure as follows, follow the tracks of results set will collect, the vortex tracker followed the tracks of of storage:
Tracker
Observation
OID Obs
Vortex ID Position detection value vector
Step 2, startup are followed the trail of, and from the next moment, temporally frame is circulated, if being recycled to last frame, chases after Track stops, and follows the tracks of the vortex tracker that results set is collected, and have recorded the Life Evolution procedural information following the trail of vortex;Otherwise, enter Row automatic tracing, performs step 3-4;
Step 3, all vortex observations of acquisition current time frame, add in observation set observeList;
Step 4, data association.Each vortex tracker in traversal tracker set trackerList, first carries out Kalman The tracking prediction of filtering, is respectively stored into vortex position and the covariance matrix of prediction in Pre and the Cov attribute of tracker, And calculate mahalanobis (geneva) distance with vortex observation position.Mahalanobis distance computing formula is as follows:
d ( X , Y ) = ( X - Y ) T Σ - 1 ( X - Y )
Wherein, X represents predictive value vector, and Y represents that observation vector, Σ represent the covariance matrix of Kalman filter.By The distribution of card side is met in mahalanobis distance, so when the predictive value vector of vortex tracker is more than with the mahalanobis distance of observation vector 0.05 significant level time, it is believed that observation prediction range of error outside, should foreclose during tracking and matching, otherwise, vortex Observation, in the range of error of prediction, is tracked mating by following several situations:
If ● be not matched to the vortex in observation set in the range of the forecast error of vortex tracker, then by this tracker from TrackerList takes out, joins in the tracking results set completeList that step 1 is set up, show the life of this vortex Evolutionary process has been followed the trail of and has been terminated;
If ● have in the range of the forecast error of vortex tracker and only one coupling vortex observation, then from vortex observation Set observeList takes out this vortex, the follow-up vortex traced into as tracker, with the renewal of the locus of this vortex with The Obs attribute of track device, is modified the position of vortex by the Kalman filter in tracker, with filtered meanwhile Vortex position and error matrix update Upd and the Cov attribute of tracker, and the process trace after carrying out of taking this as the standard;
If ● there is the vortex observation of multiple coupling in the range of the forecast error of vortex tracker, then calculate initially with figure traversal Method extracts the tracker in bipartite graph and observation, and the mahalanobis distance calculated between tracker and observation obtains distance costs square Battle array, then uses Hungarian algorithm that this distance costs matrix is carried out optimization matching.For matching the vortex of observation Tracker, takes out this vortex from vortex observation set observeList, the follow-up vortex traced into as tracker, with being somebody's turn to do The locus of vortex updates the Obs attribute of tracker, and carries out the position of vortex by the Kalman filter in tracker Revise, with filtered vortex position and Upd and the Cov attribute of error matrix renewal tracker, and take this as the standard and carry out it After process trace;
● if observation set exists the vortex of non-matched jamming device, the most newly-built vortex tracker, add to vortex with In track device set.
The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, for the technical staff of this neighborhood For, the present invention can change and change.All within the spirit and principles in the present invention, any amendment of being made, equivalent, Improve, within should be included in scope of the presently claimed invention.

Claims (2)

1. one kind based on Bayesian vortex method for automatic tracking, it is characterised in that step is as follows:
Step 1, initialization, set up an empty vortex tracker set, a vortex observation set and an empty tracking Results set;Each vortex tracker will record the follow-up pass of preamble of a vortex evolutionary process during automatic tracing System;Vortex observation set will record the eddy information that each moment observes, follow the tracks of results set by storage record life process The vortex tracker followed the trail of;The each vortex observing the first moment, builds vortex tracker, the position of record vortex Information, and create the Kalman filter of default parameters for tracking prediction, then tracker is added to vortex tracker set In;
Step 2, startup are followed the trail of, and from the next moment, temporally frame is circulated, if being recycled to last frame, chases after Track stops, and follows the tracks of the vortex tracker that results set is collected, and have recorded the Life Evolution procedural information following the trail of vortex;Otherwise, enter Row automatic tracing, performs step 3-4, until end time just end loop, stops following the trail of;
Step 3, all vortex observations of acquisition current time frame, add in observation set;
Step 4, data association, traversal tracker set in each vortex tracker, first carry out Kalman filter with Track is predicted, then by all vortexs in the vortex position of prediction and the vortex observation set of range of error and current time, carries out Association coupling.
It is the most according to claim 1 based on Bayesian vortex method for automatic tracking, it is characterised in that: described step 4 Middle data association, each the vortex tracker in traversal tracker set, first carry out the tracking prediction of Kalman filter, then By all vortexs in the vortex position of prediction and the vortex observation set of range of error and current time, it is associated coupling Process:
(41) if the vortex being not matched in the range of the forecast error of vortex tracker in observation set, then by this tracker Take out from set, join in the tracking results set that step 1 is set up, show that the Life Evolution process of this vortex has been followed the trail of Terminate;
(42) if having in the range of the forecast error of vortex tracker and only one coupling vortex observation, then from vortex observe Value set takes out this vortex, the follow-up vortex traced into as tracker, and use the Kalman filter sight to this vortex Location is put and is modified, it is thus achieved that filtered vortex position and range of error, and the process trace after carrying out of taking this as the standard;
(43) if there is the vortex observation of multiple coupling in the range of the forecast error of vortex tracker, then initially with The bipartite graph (bipartite graph) that this tracker and vortex observation are constituted by Hungarian algorithm carries out optimization matching, For matching the vortex tracker of observation, this observation is taken out from observation set, after tracing into as tracker Continue vortex, and uses Kalman filter to be modified the observation position of this vortex, it is thus achieved that filtered vortex position and mistake Difference scope, and the process trace after carrying out of taking this as the standard;
(44) if observation set exists the vortex of non-matched jamming device, the most newly-built vortex tracker, adds vortex to In tracker set.
CN201610191428.8A 2016-03-30 2016-03-30 A kind of vortex method for automatic tracking based on Bayes Active CN105894535B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610191428.8A CN105894535B (en) 2016-03-30 2016-03-30 A kind of vortex method for automatic tracking based on Bayes

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610191428.8A CN105894535B (en) 2016-03-30 2016-03-30 A kind of vortex method for automatic tracking based on Bayes

Publications (2)

Publication Number Publication Date
CN105894535A true CN105894535A (en) 2016-08-24
CN105894535B CN105894535B (en) 2019-05-31

Family

ID=57014673

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610191428.8A Active CN105894535B (en) 2016-03-30 2016-03-30 A kind of vortex method for automatic tracking based on Bayes

Country Status (1)

Country Link
CN (1) CN105894535B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107292911A (en) * 2017-05-23 2017-10-24 南京邮电大学 A kind of multi-object tracking method merged based on multi-model with data correlation
CN107657148A (en) * 2017-11-09 2018-02-02 中国海洋大学 Breakpoint based on long-time span whole world mesoscale eddies is continuous to chase after technology
CN111695299A (en) * 2020-06-04 2020-09-22 哈尔滨工程大学 Mesoscale vortex trajectory prediction method
CN112767711A (en) * 2021-01-27 2021-05-07 湖南优美科技发展有限公司 Multi-class multi-scale multi-target snapshot method and system

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080111730A1 (en) * 2006-11-09 2008-05-15 Zhen Ding Track quality based multi-target tracker
CN102148921A (en) * 2011-05-04 2011-08-10 中国科学院自动化研究所 Multi-target tracking method based on dynamic group division
CN102289594A (en) * 2011-08-19 2011-12-21 中国科学院地理科学与资源研究所 Algorithm for automatically identifying and reconstructing process of mesoscale ocean eddy
CN102737386A (en) * 2012-05-15 2012-10-17 北京硅盾安全技术有限公司 Moving target anti-fusion shielding tracking algorithm
CN102915545A (en) * 2012-09-20 2013-02-06 华东师范大学 OpenCV(open source computer vision library)-based video target tracking algorithm
CN103281476A (en) * 2013-04-22 2013-09-04 中山大学 Television image moving target-based automatic tracking method
CN103399905A (en) * 2013-07-29 2013-11-20 中国科学院地理科学与资源研究所 Method of information extraction for ocean eddy evolution process
US8705793B2 (en) * 2009-10-08 2014-04-22 University Of Southern California Object tracking by hierarchical association of detection responses

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080111730A1 (en) * 2006-11-09 2008-05-15 Zhen Ding Track quality based multi-target tracker
US8705793B2 (en) * 2009-10-08 2014-04-22 University Of Southern California Object tracking by hierarchical association of detection responses
CN102148921A (en) * 2011-05-04 2011-08-10 中国科学院自动化研究所 Multi-target tracking method based on dynamic group division
CN102289594A (en) * 2011-08-19 2011-12-21 中国科学院地理科学与资源研究所 Algorithm for automatically identifying and reconstructing process of mesoscale ocean eddy
CN102737386A (en) * 2012-05-15 2012-10-17 北京硅盾安全技术有限公司 Moving target anti-fusion shielding tracking algorithm
CN102915545A (en) * 2012-09-20 2013-02-06 华东师范大学 OpenCV(open source computer vision library)-based video target tracking algorithm
CN103281476A (en) * 2013-04-22 2013-09-04 中山大学 Television image moving target-based automatic tracking method
CN103399905A (en) * 2013-07-29 2013-11-20 中国科学院地理科学与资源研究所 Method of information extraction for ocean eddy evolution process

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
QIN WAN 等: "Multiple Moving Objects Tracking under Complex Scenes", 《PROCEEDINGS OF THE 6TH WORLD CONGRESS ON INTELLIGENT CONTROL》 *
郝菲: "智能监控系统中运动目标检测跟踪的研究", 《中国优秀硕士学位论文全文数据库•信息科技辑》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107292911A (en) * 2017-05-23 2017-10-24 南京邮电大学 A kind of multi-object tracking method merged based on multi-model with data correlation
CN107292911B (en) * 2017-05-23 2021-03-30 南京邮电大学 Multi-target tracking method based on multi-model fusion and data association
CN107657148A (en) * 2017-11-09 2018-02-02 中国海洋大学 Breakpoint based on long-time span whole world mesoscale eddies is continuous to chase after technology
CN107657148B (en) * 2017-11-09 2021-09-28 中国海洋大学 Breakpoint continuous chasing method based on long-time span global mesoscale vortex
CN111695299A (en) * 2020-06-04 2020-09-22 哈尔滨工程大学 Mesoscale vortex trajectory prediction method
CN111695299B (en) * 2020-06-04 2022-12-13 哈尔滨工程大学 Mesoscale vortex trajectory prediction method
CN112767711A (en) * 2021-01-27 2021-05-07 湖南优美科技发展有限公司 Multi-class multi-scale multi-target snapshot method and system
CN112767711B (en) * 2021-01-27 2022-05-27 湖南优美科技发展有限公司 Multi-class multi-scale multi-target snapshot method and system

Also Published As

Publication number Publication date
CN105894535B (en) 2019-05-31

Similar Documents

Publication Publication Date Title
Pang et al. Quasi-dense similarity learning for multiple object tracking
Feichtenhofer et al. Detect to track and track to detect
CN105894535A (en) Bayes-based vortex automatic tracking method
Andriyenko et al. Discrete-continuous optimization for multi-target tracking
CN103268616B (en) The moveable robot movement human body tracing method of multi-feature multi-sensor
US7026979B2 (en) Method and apparatus for joint kinematic and feature tracking using probabilistic argumentation
Andriyenko et al. An analytical formulation of global occlusion reasoning for multi-target tracking
CN106327526A (en) Image object tracking method and image object tracking system
Zhu et al. Crowd tracking with dynamic evolution of group structures
CN107590821B (en) Target tracking method and system based on track optimization
Anati et al. Robot localization using soft object detection
CN103064086A (en) Vehicle tracking method based on depth information
CN104133983B (en) Multiple-target Bernoulli distribution type fusion method based on weighting negative first-order RD (Renyi Divergence)
CN106355604A (en) Target image tracking method and system
CN103729861A (en) Multiple object tracking method
Yazdian-Dehkordi et al. Refined GM-PHD tracker for tracking targets in possible subsequent missed detections
Shakeri et al. Real-time bird detection based on background subtraction
Korthals et al. Multi-modal detection and mapping of static and dynamic obstacles in agriculture for process evaluation
Bashar et al. Multiple object tracking in recent times: A literature review
Stübler et al. A continuously learning feature-based map using a bernoulli filtering approach
Zhang et al. Toward Coordination Control of Multiple Fish-Like Robots: Real-Time Vision-Based Pose Estimation and Tracking via Deep Neural Networks.
Liu et al. Multi-sensor tracking with non-overlapping field for the GLMB filter
Yao et al. City-scale multi-camera vehicle tracking based on space-time-appearance features
Ma et al. Radiation intensity Gaussian mixture PHD filter for close target tracking
CN104240265A (en) Multi-target tracking method and system based on global restrictions

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant