Abstract: Local spatio-temporal features have become a popular video representation for action recognition in recent years.Several methods for feature detection and description have been proposed in the literature and promising recognition results are demonstrated for a number of action classes.This paper employs the motion representation based on space-time interest points and implements action recognition method based on spatio-temporal codebook and words.Firstly，accurate interests points detectes from videos taking advantage of Gabor and Gaussian mixture filtering，then three kinds of local features：histogram of gradient，histogram of flow and histogram of space-time gradient are extracted as 3DSIFT to describe interest points.K-means cluster algorithm performs on features and learns the spatial-tempoal codebook.Finally a standard bag-of-features SVM approach is used for action recognition.The performance is investigated on a total of 16 action classes distributed over two datasets with varying difficulty.Experiment results demonstrate that features combined spatial with temporal information can well adapted to complex environment such as camera movement，illumination changes and different clothing in realistic settings and achieve better recognition performance.

Key words: spatio-temporal feature, space-time interest point, motion representation, action recognition

摘要： 近年来，基于局部时空特征的运动表征方法已被越来越多地运用于视频中的动作识别问题，相关研究人员已经提出了多种特征检测和描述方法，并取得了良好的效果。但上述方法在适应摄像头移动、光照以及穿着变化等方面还存在明显不足。为此，提出了基于时空兴趣点局部时空特征的运动表示方法，实现了基于时空单词的动作识别。首先采用基于Gabor滤波器和Gaussian滤波器相结合的检测算法从视频中提取时空兴趣点，然后抽取兴趣点的静态特征、运动特征和时空特征，并分别对运动进行表征，最后利用基于时空码本的动作分类器对动作进行分类识别。在Weizmann和KTH两个行为数据集进行了测试，实验结果表明：基于时空特征的运动表示能够更好地适应摄像头移动、光照变化以及施动者的穿着和动作差异等环境因素的影响，取得更好的识别效果。

关键词: 时空特征, 时空兴趣点, 运动表征, 动作识别