A Non-Parametric Motion Model for Foreground Detection in Camera Jitter Scenes

Foreground detection in video sequences is a fundamental step of extracting information in many visual surveillance applications. However, accurate foreground detection is obstructed by the vibration of cameras. In this letter, we present a non-parametric motion model for foreground detection in camera jitter scenes. More specifically, by studying the distinction between dynamic information of unstable background and that of moving foreground in camera jitter scenes, we find that the distribution of dynamic information of unstable background is relatively constant while that of moving foreground is uncertain. Inspired by this distinction, we propose to model the distribution of dynamic information of unstable background with a non-parametric estimation technique. Then a pixel is detected as foreground if its dynamic information is different from the reference model. Experimental results indicate that the proposed method achieves better performance in foreground detection in camera jitter scenes compared with several methods in the literature, especially those detecting foreground only with color distributions.