robust and fast collaborative tracking with two stage sparse optimization
DESCRIPTION
Robust and Fast Collaborative Tracking with Two Stage Sparse Optimization. Authors: Baiyang Liu, Lin Yang, Junzhou Huang , Peter Meer, Leiguang Gong and Casimir Kulikowski. Outline. Problem of Tracking State of the art algorithms The proposed algorithm Experiment result. The problem. - PowerPoint PPT PresentationTRANSCRIPT
Robust and Fast Collaborative Tracking with Two Stage Sparse Optimization
Authors: Baiyang Liu, Lin Yang, Junzhou Huang, Peter Meer, Leiguang Gong and Casimir Kulikowski
Robust and Fast Collaborative Tracking with Two Stage Sparse OptimizationOutlineProblem of Tracking
State of the art algorithms
The proposed algorithm
Experiment resultThe problemTracking: estimate the state of moving target in the observed video sequencesChallengesIllumination, pose of target changesObject occlusion, complex background cluttersLandmark ambiguityTwo categories of trackingDiscriminativeGenerativeOutlineProblem of Tracking
State of the art algorithms
The proposed algorithm
Experiment resultRelated workMultiple Instance Learning boosting method(MIL Boosting) put all samples into bags and labeled them with bag labels.Incremental Visual Tracking(IVT) the target is represented as a single online learned appearance modelL1 norm optimization a linear combination of the learned template set composed of both target templates and the trivial template. Basic sparse representationSparse representation
Basis pursuit
DisadvantagesComputationally expensiveTemporal and spatial features are not consideredThe background pixels do not lie on the linear template subspace
OutlineProblem of Tracking
State of the art algorithms
The proposed algorithm
Experiment result
Problem AnalysisGiven ,Let , ,
Feature space can be decreased to K0 dimension
Two stage greedy method
Stage I: Feature selectionLoss function
Given , L= as labels,To minimize the loss function, solve the sparse problem below
Feature selection matrix
Stage II: Sparse reconstructionProblem after stage I
Simplify the aim function above as
Bayesian tracking frameworkLet represents the affine paramtersEstimation of the state probability prediction:
updating:
Transition model: ~ likelihood where
12Review of the algorithm
OutlineProblem of Tracking
State of the art algorithms
The proposed algorithm
Experiment result
Visual results
Quantitative results