Real time saliency detection in videos using Motion Field entropy

In this paper we propose a novel real time video sailency detection algorithm based on disorder in Motion Field. The proposed algorithm operates on the basic premise that higher disorder pertains to higher information in the scene. Based on the quantified value of the disorder, salient areas in the video frame are demarcated. In order to achieve real time operational capability, the algorithm operates in compressed H.264 domain rather than in pixel domain. The proposed algorithm has been evaluated on standard video sequences and results on real time video surveillance data are also presented.