The aim is to track visual objects from videos captured from single/multiple moving/static cameras. Some typical examples of applications are: video surveillance in school, hospital, airport, train station, electrical power sub-station and components, warehouse, home; traffic monitoring and analysis for tracking vehicles, analyzing traffic frequency, analyzing vehicles near traffic lights, or monitoring toll roads. We focus on challenging problems in tracking objects from complex video scenes: such as objects experience long-term partial/full occlusion, experience significant out-of-plane pose changes, and with complex dynamic backgrounds. Objects to be tracked are generic, can be humans, faces, vehicles, animals, workshop tools, depending on the choice of individual user. Our study focuses on the following methods:
• Local and global object appearance and shape modeling (e.g., particle filters, mean shift, and local point features);
• Differentiable manifolds (e.g. Grassmann, Riemannian) and nonlinear dynamic models;
• Domain-shift online learning and occlusion handling;
• Multi-view tracking using geometry, homography and optimal criteria;
• Fusion of thermal/near IR and visual band video information.