监控视频聚簇模式挖掘及其应用

Cluster Pattern Mining of Surveillance Video

在监控视频中,新输入的视频帧与可更新背景的差异可以实时反映监控场景中运动目标的大小和多少等信息。本文以此计算帧运动量,提出一种监控视频场景聚簇模式挖掘方法。先依据帧运动量的变化分割监控视频流,然后计算视频段的平均运动量,并对其进行K均值聚类分组,最后利用获得的聚簇知识,可以对监控视频实现多尺度摘要和相似视频检索。

In surveillance video, the difference between an input frame and an updatable background template reflects the magnitude and size of moving objects in the scene. This paper proposes an approach for mining scene clusters based on the difference from the frame motion. Firstly, the surveillance video is grouped according to the magnitude of the motion between frames. The average motion of each video segment is computed and then clustered by K-means algorithm. Finally, using the discovered knowledge, the multi-scale summarization and the retrieval of similar segments for the surveillance video can be performed.