Parametric and Non-parametric approaches for Non-Linear Tracking of Moving Objects

Our primal goal in this project is to keep track of an object of interest in a sequence of images. We assume that a camera is fixed, the target is a single solid object, and (partial) image of the target is provided. We applied two different approaches: parametric and non-parametric estimation techniques. The comparison of these approaches shows interesting results (e.g., how well they perform and what causes the filters to fail). In section (2), a general description of the Bayesian inference framework is introduced. In section (4), Extended Kalman Filter (EKF) approach is explained. Formulation of the problem and simulations are presented. Section (5) presents theoretical background and simulation result of Particle Filters (PF). In Section (6), experimental results using car traffic image sequence are shown. Finally, we conclude the report by comparing these approaches for the tracking problem.