Font Size: a A A

Background Modeling And Object Tracking Based On Particle Filter Under Complex Scene

Posted on:2010-06-28Degree:MasterType:Thesis
Country:ChinaCandidate:H H LiFull Text:PDF
GTID:2178360275976682Subject:Signal and information systems
Abstract/Summary:PDF Full Text Request
Recently, with the occurrences of terrorism attacks and the incidents of civil aviation security increasing gradually, keeping public security has been a higher demand than ever, which results in a big progress in Video Surveillance Technology(VST). In VST, making a surveillance system more intelligent, applicable and effective is a hot topic. Video processing with complicated conditions is a prominent aspect, and its results will affect high-level video understanding.This thesis tries to get insights on some key issues in video surveillance, including moving object detecting and multiple objects tracking under complicated conditions. For object detecting, firstly we introduce some sophisticated algorithms which are commonly used in many applications, and analyze their performances in different scenarios. Then, the Mixture of Gaussians (MOG) model for background modeling has been thoroughly discussed. Based on analysis of the MOG, we find it is sensitive to sudden illumination variation, and has high computational complexity. Considering the two limitations mentioned above, the mixture of Gaussians background modeling method with more constrains based on GVF-Snake model is proposed. Finally, experimental results show that the proposed method worked well in complex background modeling.For object tracking, this thesis proposes a Genetic particle filter method based on color and texture models, and experiment results demonstrate that the proposed method is more robust to clutter background. Considering object tracking with abnormal states, such as object sudden disappearance, emergence, mergence and severe occlusion, tracking maintenance approaches always have been used. Here, we propose the nearest neighbor data association and belief theory to deal with several abnormal cases in real-time. Finally, experimental results show that the proposed method is more effective under abnormal states.
Keywords/Search Tags:moving object detecting, Mixture of Gaussian model, multiple object tracking, particle filter, belief theory
PDF Full Text Request
Related items