Motion Detection Based on Twice Segmentation

Abstract

Motion detection is widely used in computer intelligent visual processing such as object tracking, object recognition and behavioural analysis. In this paper, Motion vector is adopted in order to get the accurate rough motion region. Motion vector can remove the small false motion objects. Frame difference method is conducted to get the motion matrix, and seed fill algorithm can be used to get the connectivity of the motion matrix. We can get the motion energy in the meantime. Some obvious false samples can be filtered out based on the motion vector. Considering of the regional characteristic of the motion objects in practical situation, we can set a size threshold to exclude these false motion regions. What’s more, the motion energy can be used as the measurement of motion detection in the subsequent refine segmentation stage. Twice segmentation is needed to obtain the accurate motion object. This method can be efficient with less noisy point and better object skeleton with shadow removal. To confirm the effectiveness of the proposed method, we tested the method in different scene and background environment.