Hybrid particle filtering for real time object tracking

This paper presents a method for real time object tracking. The method tracks 3D objects in image sequences and jointly estimates their 3D pose and motion parameters. The solution relies on a state modeling of this estimation problem. We develop a resolution method based on a sequential Monte Carlo method and more particularly on a hybrid particle filter. This approach combines the benefits of the linear filtering with those of the nonlinear filtering by using the linear part of state equations. The proposed method allows a significant reduction in running time and preserves the optimality of the processing. As a consequence, the proposed method allows a real time object tracking