Abstract: Traditional gesture recognition methods always focus on hand contours or hand movement track,and the extracted gesture features are often difficult to represent the difference between dynamic gestures accurately.To overcome the issues of complex time series,the spatial variability and inaccurate feature representation of different dynamic gestures,a novel dynamic gesture recognition method is proposed here by combining global gesture motion and local finger motion.Firstly,based on the given hand joint positions,several data pre-processing steps are performed for dynamic gesture data,such as removing of the invalid gesture frames,completing the gesture frames,and the normalization of joint lengths for different gestures.Secondly,the key gesture frames will be extracted according to the distance function defined by the difference of hand translation and rotation,fused by the difference of panning and rotating of fingers.Meanwhile,according to the extracted key gesture frames,the gesture features of global gesture motion and local finger motion can be calculated.Finally,by combining the extracted gesture features,dynamic hand gestures can be classified and recognized using linear discriminant analysis(LDA)and Gaussian kernel based SVM.The proposed method has been evaluated on the DHG-14/28 datasets,which includes 14 kinds of gestures and 28 kinds of gestures.And the accuracy of hand gesture recognition is 98.57%and 88.29%respectively,which is 11.27%and 4.89%higher than the existing methods.Experimental results demonstrate that our method can represent the difference between dynamic hand gestures accurately and recognize them effectively.