Real-time distributed video tracking of multiple objects from single and multiple cameras

Multiple object tracking in video has received tremendous attention because of its wide practical applications such as video surveillance, human activity analysis, target identification, and human computer interfaces.; Due to the complexity associated with high dimensionality, occlusions, motion changes and background clutters, robust and efficient video tracking of multiple objects remains a challenging task. Multiple independent single object trackers fail when objects are in close proximity or present occlusions. In such circumstances, modeling the interaction among objects, establishing a correspondence between objects and observations, and decreasing the computational complexity to achieve real-time implementation are critical problems.; In this dissertation, we investigate issues towards solving these problems. Specifically, the thesis comprises five fundamental contributions: The first is a Detection-Based Particle Filter, which extends the particle filter theory and achieves robust performance of single object tracking. Secondly, a distributed Bayesian formulation is proposed for real-time multiple object tracking using a single camera. It avoids the common practice of using a complex joint state space representation and performing tedious joint data association. It extends the conventional Bayesian tracking framework by modeling multiple object interaction in terms of potential functions. The third contribution is a distributed framework using multiple collaborative cameras for multiple object tracking with significant and persistent occlusion. Specifically, we propose to model the camera collaboration likelihood density by using epipolar geometry with sequential Monte Carlo implementation. Fourthly, we have proposed two novel approaches for articulated object tracking. Instead of using a high dimensional joint state representation, we introduce a decentralized scheme and model the inter-part interaction within an innovative framework. Finally, we present a novel video tracking framework using control-based observer design. It unifies several kernel-based approaches into a consistent theoretical framework by modeling tracking as an inverse problem. It relies on observability theory from control systems to handle the "singularity" problem and provides explicit criteria for kernel design and dynamics evaluation.