A novel method of surveillance video Summarization based On clustering and background subtraction

Abstract

In this paper, a novel surveillance video summaryzation approach is proposed to detect the objects and targets which appear less frequently. This approach integrates clustering and background subtraction. The clustering method adopts a modified Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to summarize the surveillance video with HSV color feature. Then the background frame is extracted from the clusters, the background subtraction method is used to generate the key-frame sequences. Finally, video summarization is got by combining the above results. Experimental results show that the Recall Ratio (RR) has a great increase comparing with one of the two methods.