A Novel Object Tracking Algorithm Based on Particle Filtering on the Affine Group

Abstract

In this paper, a novel approach based on particle filtering on the affine group is introduced for object tracking. We firstly use Scale invariant feature transform (SIFT) to extract corresponding feature points between two successive frames. The affine parameters for pose estimation from the corresponding feature points can be formed as a solution to Sylvester's equation. Then, we can smoothly estimate the affine parameters within the particle filtering framework. Where the state dynamic is modeled via the first order autoregressive (AR) process on the affine group, at the same time, the optimal state of the tracked object is estimated through the total likelihood function using appearance model and feature model. Experimental results demonstrate the robustness and efficiency of our proposed approach for object tracking.