Moving Objects Representation for Object Based Surveillance Video Retrieval System

Increasingly large amounts of surveillance video data have resulted in the critical need for indexing and retrieval from video databases. This paper addresses the problem of objects representation for surveillance video retrieval system. We extract dominant color histogram and edge direction histogram as the object's appearance model in which the improved histogram intersection algorithm is used to measure the similarity between two dominant color histograms. Furthermore, an agglomerative hierarchical based clustering method is proposed to select the most relevant and representative blobs for compact and effective object representation. Experimental results on real surveillance video sequences have proved the performance of our proposed approach.