As an effective means of human-computer interaction, gesture recognition has become a hot topic in current research. In order to solve the problems of spatio-temporal variability and feature complexity concerning dynamic gestures, we propose a dynamic gesture recognition solution based on 3D skeleton features. The accuracy of dynamic gesture recognition is greatly impaired due to the temporal differences and complexity of dynamic gestures, thus a key frame extraction algorithm is designed to extract key features of dynamic gestures for further feature extraction. To overcome the difference in classification performance between single classifiers, multiple classifiers are used to simultaneously classify and fully exploit gesture features. We also propose an adaptive fusion algorithm to effectively fuse different classifiers according to their classification performances thus improve the final classification accuracy. Finally, experiments are carried out, and results verify the effectiveness of the proposed dynamic gesture recognition approach.