Tracking Objects in Nonstationary Environment

Abstract

We aim at tracking objects in a nonstationary environment. Object tracking can be a simple task when the background is stationary and the subtraction method can be applied to obtain the object shape. However, the method is infeasible when the background is changed frequently, like waves, moving clouds or tree branches which are usually not the objects that we care about. We propose a method applying mixture models, with a Markovian assumption to detect the foreground object. The method is operated in a pixelwise manner, followed by a spatial smoothing. After a certain training period, the built hidden Markov model is adopted for subsequent real-time object detection. This is a joint work with Lung-An Li, Yu-Hsiang Lee and Ching-Hau Chen.