中文摘要：

针对目前局部描述方式中忽略邻点之间关系从而丢失邻域结构信息,以及忽略时空特征在时域和空域上变化具有差异性的问题,提出一种基于时空兴趣点的时空纹理特征描述方法 DT-LBP-TOP。实现邻域中方向信息的描述,解决变化率以及变化方向的问题,实现针对时域特点的邻域像素间关系的描述,对时域和空域特征进行区分描述,获得更加丰富且具有区分度的描述信息。实验结果表明,该算法有效描述了特征信息,行为识别精度高达96.5%。

英文摘要：

The current local description methods ignore the relationship between adjacent points and lose the neighborhood structure information,and they neglect the problem of spatial and temporal changes in time domain and spatial domain,a spatio-temporal texture feature description method based on spatio-temporal interest points was proposed.The description of the direction information in the neighborhood was realized,and the problem of the change rate and the change direction was solved.The description of the relationship between neighboring pixels in time domain was realized,and a distinction between the different changes of the temporal and spatial characteristics was made.Experimental results show that the proposed algorithm can effectively describe the feature information,and the behavior recognition accuracy is up to 96.5%.