Motion Flow Clustering for Moving Vehicle Detection from Satellite High Definition Video

Current traffic monitoring is limited by the small coverage of camera surveillance systems, for example, a specific area around one road intersection. Satellite high definition videos are becoming available which can provide videos over several squared kilometers. Thus, these videos introduce new possibilities for better traffic control and management. However, parallax motions caused by the movements of the satellite platform make accurate moving vehicle detection from these videos a challenging problem. In this paper, we propose a novel approach to remove the effects of parallax motions in moving vehicle detection from satellite high definition videos. Motion flows are gathered after candidate pixels are extracted using a background subtraction method, and then a motion flow clustering method is employed to gather motion patterns from motion flows. By utilizing the dissimilarity in motion patterns between moving vehicles and parallax, false detection candidates can be removed effectively. In this paper, both qualitative and quantitative evaluations were conducted. Experimental results demonstrate that our approach outperforms existing approaches in terms of precision.