Moving object boundary extraction using level set method
School of Electrical and Electronic Engineering
This thesis deals with the problem of automatic extraction of moving object boundary from video data. Since the level set method plays an important role in object boundary extraction, this thesis starts with a review of this method. Various models based on the level set method are presented and their advantages and disadvantages are discussed. For a level set based method to be useful in automatic boundary extraction, two major considerations must be taken into account; 1) the formulation of an energy functional that provides desired features and 2) its high computational cost. To address these issues, the research work in this thesis focus on developing a fast level set based object boundary extraction algorithm. A region based energy functional based on background subtraction and Chan-Vese model is proposed. This energy functional has the advantage of having robustness to handle noisy and cluttered environment compared to the gradient based energy functional. A way to incorporate this energy functional into a fast level set implementation scheme is also introduced. With this scheme, level set based curve evolution can be performed in a faster way compared to the existing PDE solver based methods. A new method to construct a shape model is proposed. The shape model is constructed based on local nearest neighbors. An energy functional based on the constructed shape model is developed and incorporated into a region based energy functional. This extension allows recovering the shape of partially occluded objects. Shape prior modeling based on principal component analysis is also implemented for comparisons with the proposed method. Experimental results demonstrate the superior performance of the proposed method in various environmental conditions.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems