Dissertation > Excellent graduate degree dissertation topics show

Research on Object Locating and Tracking for Visual Sensor Network

Author: ZhouLiangYi
Tutor: WangZhi
School: Zhejiang University
Course: Control Theory and Control Engineering
Keywords: objects locating and tracking feature fusion Bayesian theory particle filter sparsity representation dynamic model
CLC: TP391.41
Type: Master's thesis
Year: 2012
Downloads: 96
Quote: 2
Read: Download Dissertation

Abstract

In the field of computer vision, locating and tracking of moving objects are both hot topics. The locating and tracking of objects are realized by image processing technology and filter technology in signal process. Traditional object locating and tracking algorithms are based on single view. But in the real scenario, there is so much noise that the detection of object is difficult. Single camera have limited FOV(field of view), occlusion between objects may introduce wrong information. With the development of VSN(Visual Sensor Network), more and more systems fuse the information from multiple views and track object based on VSN.There are still many problems need to be solved. For example, object feature is not stable, the object size in image plane is not stable, there will be occlusion between objects. In order to locate and track objects with high precision. We need to design a good fusion algorithm based on multi-view and improve the algorithm of locating and tracking. This dissertation does some research on object locating and tracking algorithm based on VSN. The main work of the dissertation are as follows:1) For the feature fusion in object tracking, an object feature fusion algorithm based on occlusion variable is proposed. This algorithm firstly introduces occlusion variable to describe the occlusion state between objects. The occlusion variable is calculated and updated with calibration information in VSN. A dynamic threshold is introduced to calculate and update the occlusion variable, because the size in image plane is different with different object locations. With occlusion variable, object features are fused on the common plane. The fusion feature is more stable than traditional feature. The occlusion variable and fusion feature are combined and introduced under the framework of Bayesian theory. Object tracking is realized with the particle filter tracking algorithm based on occlusion variable and fusion observation model. The results show that features fusion algorithm and tracking algorithm we proposed can solve the problems including the change of object scale, stability of features and occlusion between objects. Objects can be tracking correctly.2) For the problem of massive information interacting among the nodes in VSN, we propose a sparsity based objects locating algorithm. In this algorithm, object feature is represented by sparsity dictionary. In order to represent the location with sparsity, the common plane is divided into different parts. With the presentation of sparsity dictionary, each location is matched with the part of the sparsity dictionary and the problem of "ghost zone " in traditional location algorithm is solved. For the difficult occlusion problem, the overlapping part of sparsity dictionary is quantized and the wrong information is removed by quantity. In order to compensate the wrong information caused by occlusion, information from multi-view needs to be fused. The locating problem is recasted as an inverse problem with sparsity. We proposed an object tracking algorithm based on sparsity feature and color feature. Based on the feature of the inverse problem in object locating, We propose a solution algorithm with object dynamic model. The results show that locating algorithm with sparsity can find objects location even when there are occlusions between objects.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile