Abstract

This thesis presents an integrated approach to modelling, extraction and tracking of deformable contour meshes through image sequences, with the aim of extracting motion information about the viewed scene. The thesis begins by reviewing the area of motion estimation in computer vision, leading to a review on the formulation and initialisation of active contour models. From this review the thesis develops and provides as its major contribution an active mesh structure that may be used for motion estimation. This active mesh structure approach is combined with feature matching to provide a stable, deformable motion tracking system for real-world scenes. This system is tested on various real-world scenes and varying conditions to provide extensive and rigorous experimental proof of the validity of the formulation. Further extensions to the system are implemented, including the use of multiple and region based active meshes. Future directions of research are also suggested.