STUDY OF ASSOCIATIONS AMONG VARIOUS VIDEO SEQUENCES

Abstract

Visual tracking has been investigated extensively in the last two decades due to its myriad of applications in the field of computer vision. It has been an active area of research due to lack of adequate holistic tracking systems that can handle uncontrolled environment scenarios. Multicue tracking solutions aim at robust and accurate estimation of object trajectory using multiple features (cues) extracted from single-modal or multi-modal data. Although, multicue object tracking systems were explored by some researchers, it still remains a challenge to fuse multicue in a probabilistic tracking framework. newlineIn this thesis, research work has been focused on recursive bayesian estimation for multicue object tracking. Under this, unknown probability density function of target state was estimated in a recursive process by extracting multicue data from video sequences. The fundamental problems in visual tracking were addressed through development of multicue probabilistic tracking frameworks and design of multicue fusion models for robust visual tracking. Considering problems of recursive Sequential Monte Carlo tracking framework, an evolutionary particle filter based on improved cuckoo search was proposed for solving sample degeneracy and impoverishment problem. Improved Cuckoo Search (ICS) algorithm was embedded into particle filter (PF) framework. ICS algorithm used levy flight for generating new particles in the solution and introduced randomness in samples by abandoning a fraction of these particles. Also, a novel state model for catering object scaling and rotational error was designed and integrated to propose tracking framework. newline