Abstract: Compared with traditional light cameras, the depth sensor has advantages of synchronous acquisition of RGB image and depth image and provides a new way for human action recognition. Thus, it uses the optimized STIPs algorithm to extract STIPs for color and depth image sequence separately. Considering the limitation of color and depth image, it integrates the STIPs according to the rules. Owing to the redundancy of combining features, spatial-temporal clustering method is used to process the STIPs and SVM classifier trains and tests the samples. The experimental result demonstrates that the result based on proposed method is as accurate as 91%, which is much better than other methods.

Key words: depth sensor, Spatial-Temporal Interests Point（STIP）, feature combination；action recognition

摘要： 与传统光学相机相比，能同步获取RGB图像和深度图像数据，对人体行为识别提供了新的解决方案。因此，分别对RGB和深度图像序列提取改进的时空兴趣点特征，并基于一定规则实现时空兴趣点特征的融合。由于融合后特征的冗余性，基于时空聚类的方法，对特征进行优化处理，并采用SVM分类器进行训练和测试。实验结果表明提出的RGB和深度图像特征联合方法的行为识别平均准确率为91%，相对于其他方法取得了更好的识别结果。

关键词: 深度传感器, 时空兴趣点, 特征融合, 行为识别