Controllable Visual Cues

This project concerns the development of analytical and computational tools for the analysis of dynamic visual processes. In particular, emphasis has been given to the estimation of visual motion for the purpose of control, the estimation of shape from moving images, the estimation of spatio-temporal image statistics for visual modeling, synthesis and recognition. The project is in its final year, and the progress has been documented extensively in the following publications [33, 36, 8, 29, 41, 32, 7, 25, 9, 40, 28, 20, 48, 22, 37, 1, 15, 2, 42, 47, 13, 45, 16, 24, 27, 30, 52, 44, 19, 12, 10, 18, 3, 39, 11, 4, 35, 17, 33, 5, 6, 14, 23, 21, 31, 49, 50, 46, 26, 38, 51, 53, 43, 34]. These include a textbook [34], a Marr Prize-winning manuscript [33] and a Siemens Prize (best paper award at CVPR) manuscript [5], and the first real-time demonstration of a system to estimate 3D structure and motion in real time on commercial off-the-shelf hardware [20]. Many of the results obtained during the course of this project have been incorporated in a new first year graduate course developed at UCLA (CS268, Machine Perception), as well as in a three-day short course outreach to industry and government titled “3D reconstruction from Video” administered through the UCLA extension in the summer of 2002 and 2003.