Surveillance key frame extraction based on aggregation dispersion entropy and moving target detection

Key frame extraction is an important step in surveillance video retrieval. We propose a surveillance key frame extraction algorithm which is based on the aggregation dispersion entropy and moving target detection. Firstly, the concept of the aggregation dispersion entropy was defined to distinguish the presence of moving objects in video. Secondly, the aggregation dispersion entropy was used to divide surveillance video into several shots. And then the shots were splitted into sub-shots by the moving target detection. So the key frames could be got though the sub-shots. Finally, the algorithm of key frame extraction was given. The experimental results and their discussions were given; they showed that this algorithm has good performance both in accuracy and robustness for several different databases. Also, it is the demand of surveillance video retrieval in police use. And it is expected to be of further use in police video investigation.