Surveillance Video Sequence Segmentation Based on Moving Object Detection

Increasing surveillance video makes it unmanageable for human to find useful information from large data collections. Video indexing and retrieval technology is one of the effective solutions, in which video sequence segmentation is a fundamental step. In this paper, a surveillance video sequence segmentation method is proposed, which utilizes foreground from moving object detection as features. First, moving objects are detected by adaptive background subtraction technique with a three-frame differencing, from which foregrounds are extracted. Then, the sum of foreground pixels is calculated as difference feature. Last, a sliding window based adaptive threshold is proposed to determine the segment boundaries. The proposed algorithm has been evaluated by seven videos, which were captured at campus roadsides and downloaded from internet. The experimental results show the algorithm is effective and efficient.