Automated surveillance systems require accurate scene segmentation, target detection and tracking, and reliable event understanding. I am particularly interested in addressing the difficult issues of robust tracking of multiple and possibly overlapping targets in the presence of occlusion, cooperation between multiple cameras, and scenario recognition based on detected trajectories.

In the area of robot vision, I am interested in developing systems that allow for autonomous navigation through complex and dynamic environments, and for object detection and manipulation. Such systems must have the ability to recover ego-motion, infer 3-D shape from motion, and recognize objects of interest in a cluttered scene. A central issue is developing robust methods for incremental structure inference, by continuously refining the description as new images are acquired.