Particle Filter Based Pose and Motion Estimation with Non-Gaussian Noise

Pose and motion estimation from a monocular sequence of images is crucial for many robot vision tasks, and particle filtering (PF) which is based on sequential importance sampling (SIS) has drawn much attention in recent years due to its capacity to handle nonlinear and non-Gaussian dynamic problems. In this paper, given a sequence of two-dimensional (2D) monocular images of an moving object, using line features on the image plane as measured inputs and a dual quaternions to represent the three-dimensional (3D) transformation, the indirect measurement solutions of pose and motion is presented based on extended Kalman filtering (EKF) and PF with simulated data. Simulation results with Gamma noise have shown that PF has good convergence, while the EKF is divergent.