摘要
提出了一种基于时空3D-sift特征和潜在主题分布的高效异常行为分析方法。该方法首先利用时间轴上的Gabor滤波器以及3D-sift特征描述子提取视频关键字。将视频信号看作文本,则关键字就是里面的单词,pLSA算法假设在单词与文本之间存在潜在的主题,根据视频中主题的分布进行异常行为分析,而不是直接采用单词分布。针对使用Gabor滤波器提取感兴趣点产生的一些冗余点进行排除,采用并行算法降低运算时间。
An abnormal behavior analysis method based on efficient space-time 3D-sift features and latent topics is proposed. This method extracts interest points by using the Gabor filters and 3D-sift video feature descriptor, and views the video as the text, thus the interest points are the words in the text. Assuming that some latent topics exist in between the words and the text, the pLSA algorithm implements abnormal behavior analysis in accordance with the topics distribution in the video rather than the words. The redundant points extracted by the Gabor filter are excluded, and also the parallel algorithms adopted to reduce the computing time.
出处
《通信技术》
2012年第7期67-71,共5页
Communications Technology
基金
国家自然科学基金(批准号:61071173)
关键词
异常行为
时空特征
PLSA
潜在主题
并行计算
abnormal behavior
space-time feature
pLSA
latent topic
parallel algorithms
