Tracking of feature points in image sequence by SMC implementation of PHD filter

We investigate a method for filtering of feature points´ trajectories in image sequence by using a novel technique named sequential Monte Carlo (SMC) implementation of probability hypothesis density (PHD) filter. PHD filter uses finite random set (FRS) on state space to represent and to track multiple targets in clutter. It can deal with appearance/disappearance of target due to the FRS representation. PHD is 1st order moment of finite random set, which corresponds to mean vector of the Kalman filter in continuous variable state case. SMC implementation of PHD filter is an elaborated filter that approximates the PHD by many number of realization, which are called particles, and it properly control the number of particles according to appearance/disappearance of targets. We apply this idea to track trajectories of feature points in image sequence. Simulation and real image analysis show the efficiency of the method.