基于Kinect的中国手语识别

基于微软Kinect提取的深度图像信息，提出了一种新的中国手语识别方法。该方法首先利用Kinect获取人体主要骨骼的3D坐标和手的3D坐标；然后根据中国手语的手型、手的位置和手的方向3个主要构造成分，分别采用DBSCAN和K-means聚类算法获取手语特征中的手的位置基元和方向基元，提出一种结合CLTree和Attribute bagging聚类集成方法提取手型基元；最后将这3类基元进行组合采用模板匹配方法识别中国手语。通过对选取的72个中国手语进行识别实验，平均识别率为90.35%，实验结果说明了方法的可行性。

Chinese sign language recognition based on the Kinect

A novel recognition method for Chinese sign language based on depth information extracted from the Microsoft Kinect is proposed in this paper. At first, the 3-D features of skeletons of mainbody and palms would be extracted based on Kinect. Secondly, the Chinese sign language can be seen composed by three components, that is, hand shape, location and hand orientation. The DBSCAN and K-means algorithms are used to extract the location subwords and orientation subwords respectively, a cluster ensemble method combined CLTree and Attribute bagging is proposed to extract the shape subwords. At last, the three subwords are combined together, and template matching method is used as a recognition method. The experimental results show that the average recognition rate is 90.35% on 72 Chinese signs, and the proposed method is proved to be effective.