Efficient video segmentation algorithm for real-time MPEG-4 camera system

Abstract

Video segmentation is one of the fundamental technologies for content-based real-time MPEG-4 camera system. For real- time requirement, the segmentation algorithm should be fast and accurate. Almost all existing algorithms are computationally intensive and not suitable for real-time applications. In this paper, based on change detection and background registration, a new segmentation algorithm is proposed. The algorithm includes two modes: first, the baseline mode that can deal with general sequences; second, the shadow cancellation mode, which is based on gradient operation and block-based changed detection, can deal with sequences affected by shadow and light change. Experimental results show that this algorithm can be applied on different kinds of sequences, and can quickly give accurate segmentation results.