Tracking video target via particle filtering on the lie group normal distribution

Most existing video target tracking algorithm based on particle filtering are in Euclidean space. While video target tracking occurs in complex environments, it is very difficult to guarantee the tracking effect. This paper describes the covariance descriptor to represent the object image region, and describes the geometric deformation of the object image region by an affine transformation. The affine transformation matrix is one element of the Lie group. We directly implement the video tracking system state lies in a low dimensional manifold, and make full use of state space of intrinsic geometrical characteristic, the manifold of optimization algorithm to solve Riemann mean value is studied. By constructing Lie group normal distribution as the optimal importance function for extracting state sample. This paper proposes particle filtering on the Lie group with optimal importance functions, which provides a kind of new train of thought for improving efficiency and robustness of the tracking algorithm. The simulation experiments show that in the case of object scale size changing, rotating, etc. geometric deformation light intensity changing, target occlusion and fast motion situation, the proposed Lie group particle filtering algorithm can still realize target tracking well and improve the real-time performance of the system.