Online video synopsis of structured motion

详细信息    查看全文

• 作者：Wei FuAuthor Vitae ; Jinqiao Wang ; ^{jqwang@nlpr.ia.ac.cn" class="auth_mail}Author Vitae ; Liangke GuiAuthor Vitae ; Hanqing LuAuthor Vitae ; Songde MaAuthor Vitae
• 关键词：Video synopsis ; Motion structure
• 刊名：Neurocomputing
• 出版年：5 July, 2014
• 年：2014
• 卷：135
• 期：Complete
• 页码：155-162
• 全文大小：1552 K

文摘

With the explosive growth of surveillance video data, video synopsis technology is presented for fast browsing a day鈥瞫 worth of video in several minutes. However, for most existing solutions, motion structure in original videos may be destroyed even considering the temporal consistency of related objects. To maintain the important context cues, in this paper, we propose an online motion structure preserved synopsis approach, which can preserve behavior interactions between different objects in the original video while condensing as much content as possible. By measuring sociological proximity of moving objects, we introduce motion structure as a refined term directly added to the problem of energy minimization. A hierarchical fashion is employed to efficiently search an optimal solution for the problem of video synopsis, in which both the spatial collision and the temporal consistency are considered. Experimental results on extensive videos demonstrate the promise of the proposed approach.