基于复合运动特征的视频异常事件检测

Hybrid motion feature based abnormal event detection for video

为提高异常检测算法的性能,提出一种基于复合运动特征和一分类模型的异常事件检测算法。为能够对视频内容进行更全面描述,提出融合一阶和二阶运动的复合运动特征,对视频内容进行全面描述;利用运动光流场去除冗余信息,只对感兴趣区域进行检测,以降低计算代价;使用一分类线性判别分析算法实现异常事件检测。在公共数据集UCSD上的实验结果表明,该方法在检测性能上优于对比方法,能够有效实现视频中的异常事件检测。

To improve the performance of abnormal event detection methods,an abnormal event detection method based on hybrid motion feature and one-class classification model was proposed.To describe the context of videos more comprehensively,the hybrid motion feature which combined the first-order and second-order motion was proposed.The redundant information was removed using the optical flow field,and only the region of interest was used for anomaly detection,to reduce the computation cost.A one-class linear discriminant analysis algorithm was used to detect abnormal event.Experimental results on public dataset UCSD show that the proposed method is superior to the compared methods and achieves satisfactory performance.