摘要
现有的视频浓缩方法普遍存在一个问题,即浓缩视频画面中包含的目标太多,不适合审阅。从视频浓缩效果的评价标准出发,提出了更合适的评价标准,改进了优化目标函数;即期望更少重叠、更少时序混乱的情况下,期望同时出现的目标数为5~9个。另外,采用差分进化算法搜索近似最优的轨迹排布。实验表明,改进后的算法有效解决了上述问题。
There is a common problem in the existing video synopsis methods,that is,the screen of the concentrated video contains too many targets,not suitable for review. Based on the evaluation criteria of video synopsis,this paper put forward a more appropriate evaluation criterion and improved the objective function; namely,the expected number of targets that are expected to occur at the same time is 5-9,with less overlap and less sequence confusion. In addition,the paper used differential evolution algorithm to search the approximate optimal arrangement of tracklet sets. The experimental results show that the improved algorithm can solve the above problem effectively.
引文
[1]Making a Long Video Short:Dynamic Video Synopsis[C].CVPR'06,Jun.2006:435-441.
[2]Pritch Y,Rav-Acha A,Peleg S.Nonchronological video synopsis and indexing[C]∥IEEE Transactions on Software Engineering,2008:1971-1984.
[3]Fu W,Wang J,Gui L,et al.Online video synopsis of structured motion[J].Neurocomputing,2014,135(8):155-162.
[4]Vural U,Akgul YS.Eye-gaze based real-time surveillance video synopsis[J].Pattern Recognition Letters,2009,30:1151-1159.