基于视频图像块模型的局部异常行为检测

Local abnormal behavior detection based on video image block

为提高视频监控下局部异常行为检测的准确性和快速性,文中提出一种基于视频图像块模型的局部异常行为检测方法。该方法首先进行基于梯度直方图的视频图像块提取和基于时空相关性的视频图像块优化;然后,对视频图像块进行基于张量投票的正常行为模型学习和基于时空相似性的异常行为模型学习;最后,依据学习结果进行视频图像块的局部异常行为检测,确定异常区域并进行有效的标记。在通用UCSD数据集和Subway数据集上的实验结果表明,文中所提方法能有效提高局部异常行为检测的准确性和检测的效率。

To improve the accuracy and the speed of the local abnormal detection, a novel method based on video image block model is proposed. Firstly the video block is extracted based on the gradient histo- gram and optimized based on the temporal-spatial coherence. Then, the normal behavior model is learned based on the tensor voting algorithm and the abnormal behavior model is learned based on the temporal- spatial coherence. Finally, the abnormal behavior is detected and labeled. The experiment is conducted on public UCSD and Subway datasets. Experimental results demonstrates that the proposed method can improve the accuracy and the efficiency of the local abnormal behavior detection.