Adaptation of Unscented Particle Filter for visual tracking in electro-optic theodolite

In this paper, we present a modified Unscented Particle Filter (UPF), and a visual tracking algorithm based on the modified UPF in electro-optic measuring system. We modify the computing methods in the UPF algorithm according to the electro-optic measuring system, which lead to a considerably low computational-complexity. We replace the standard σ sampling in UPF with a procedure of matrix decomposition and matrix reconstruction. We simplified the nonlinear transform of σ sampling, state vector, measurement vector and their covariance matrix in the standard UPF using an MSUPF (UPF for Mixed system) algorithm, where the state model is linear, and the observation model is nonlinear in the electro-optic measuring system. Experiment result shows that the tracking accuracy of the proposed MSUPF is better than that of the standard UPF, and the proposed MSUPF algorithm consumes only about 81.13% computation time of the standard UPF.