Unified framework for object tracking and recognition based on condensation principal component analysis in a structured environment

Abstract

In recent years, computer vision research has witnessed a growing interest in subset analysis techniques. In particular, eigenvector decomposition has been shown to be a highly effective tool for problems which has high-dimensional signal formats (e.g., an image array) but, nevertheless, represent visual phenomena which are intrinsically low-dimensional. Subspace analysis is heavily used in appearance-based modelling and recognition where the principal modes or the characteristic degrees-of-freedom are extracted and used for description, detection, and recognition. The complex nonlinear appearance manifold expressed as a collection of subsets, and the connectivity among them. The connectivity encodes the transition probability between images in each manifold and is learned from a training video sequences. When we track and recognize the object, a single frame image is used for that tasks. In this case based on PCA, the undesired classification/recognition results often occur. In this thesis, Condensation PCA (CPCA) presentation is introduced, which can be used for spatio-temporal alignment in tracking and recognition tasks.