Dynamic gesture recognition algorithm Combining Global Gesture Motion and Local Finger Motion for interactive teaching

Abstract

Adding dynamic gesture recognition to the new interactive teaching system is of great significance to improve the teaching efficiency. However, the features extracted by traditional dynamic gesture recognition methods are usually difficult to accurately represent the differences between dynamic gestures. Aiming at the problems of complex time sequence and spatial variability of dynamic gestures, this paper proposes a gesture recognition method combining global motion and local motion of fingers. Firstly, preprocessing of dynamic gesture data is carried out, including removing invalid gesture frames, completing gesture frame data and normalizing joint length. Then, according to the given hand joint coordinates, dynamic gesture key frames are extracted by using gesture distance function, and the global motion features of hand in space and the local motion features of fingers in hand are extracted based on gesture key frames. Secondly, the key frame gesture features of gesture global motion and finger local motion are fused, and linear discriminant analysis is used for feature dimensionality reduction. Finally, support vector machine with Gaussian kernel is used to realize dynamic gesture recognition and classification. Experimental results on DHG-14/28 and FPHA dynamic gesture dataset show that the accuracy of classification and recognition is 98.57% 88.29% and 97.31% respectively.