Real-Time Foreground Segmentation from Moving Camera Based on Case-Based Trajectory Classification

Abstract

Recently, several methods for foreground segmentation from moving camera have been proposed. A trajectory-based method is one of typical approaches to segment video frames into foreground and background regions. The method obtains long term trajectories from entire of video frame and segments them by learning pixel or motion based object features. However, it often needs large amount of computational cost and memory resource to maintain trajectories. We present a trajectory-based method which aims for real-time foreground segmentation from moving camera. Unlike conventional methods, we use trajectories which are sparsely obtained from two successive video frames. In addition, our method enables using spatio-temporal feature of trajectories by introducing case-based approach to improve detection results. We compare our method with previous approaches and show results on challenging video sequences.