An approach for fast and parallel video processing on Apache Hadoop clusters

This paper proposes an approach for fast and parallel video processing on MapReduce-based clusters such as Apache Hadoop. By utilizing clusters, the approach is able to handle large-scale of video data and the processing time can be significantly reduced. Technique details of performing video analysis on clusters are revealed, including method of porting typical video processing algorithms designed for a single computer to the proposed system. As case studies, face detection and motion detection and tracking algorithms have been implemented on clusters. Performance experiments on an Apache Hadoop cluster of six computers show that the system is able to reduce the running time of the two implemented algorithms to below 25% of that of a single computer. The applications of the system include smart city video surveillance, services provided by video sites and satellite image processing.