视觉关注转移的事件检测算法

摘要

智能监控系统已广泛应用于银行、超市、公交车等公共场合,监控视频的事件检测已经成为智能监控中的关键技术.提出了一种基于视觉关注转移的事件检测方法,该方法首先分别通过对视频帧进行动态和静态受关注模型的提取得到视觉关注显著图,然后根据视觉关注显著图的时域特性形成视觉关注节奏曲线,根据视觉关注节奏的变化强度选取关键帧,以关键帧形式表示受关注事件的发生.实验结果表明,算法提取的关键帧可以准确地标示监控视频中特征事件的发生,并且可以做到实时地检测事件.%Intelligent surveillance systems have been widely used in banks, supermarkets, buses, and other public places. Event detection in a surveillance video is a key technology for this field. In this paper, a visual attention shift-based event detection algorithm was proposed for intelligent surveillance in which the dynamic and static visual attention regions were detected to obtain the visual saliency map. After that, the visual attention rhythm was derived from the visual saliency map temporally. According to the visual attention rhythm, the key frames were selected to label the occurrence of the events. Experimental results demonstrate that the proposed algorithm can label the oc-currence of the events with the extracted key frames correctly, and that the event detection is performed in real-time.