Video Processing and Analysis

This course introduces methods to acquire state (both spatial and temporal) estimations from video streams. Video streams are analyzed as dynamic systems, linear and non-linear. If the system can be approximated as linear and Gaussian in terms of the dynamic noise in the process and measurement, then Kalman filter techniques are used. This refers to sequential state estimation. For nonlinear dynamical systems, the EFK (Extended Kalman filter) can be used by a liberalization of the stat space model. Particle filters are used to address state estimation problems where the systems are non-linear and non-Gaussian. Particle Filters are rooted in Bayesian estimation and Monte Carlo procedures. The course builds upon these techniques in studying visual tracking and the components of Visual SLAM (Simultaneous Localization and Mapping) procedures.