Gesture Recognition Algorithm Based on Lightweight 3DCNN Network
Guoping Zhang, Nan Ma, Jiahong Li, Beiyan Jiang, Zhixuan Wu · 2021 17th International Conference on Computational Intelligence and Security (CIS) · 2021
Aiming at the slow recognition speed of the video- based dynamic gesture recognition method and the large number of parameters in the deep 3D CNN network, we propose a fast gesture recognition method, which is an improved lightweight network Lite-HRNet. Firstly, we convert the 2D lightweight network Lite-HRNet to a 3D CNN network to recognize video-based dynamic gestures while generating fewer parameters. Secondly, this work only converts the first half of the Lite-HRNet network to a 3D CNN network, which can significantly improve the recognition speed of the method. We extensively evaluate our method on the EgoGesture gesture dataset. Experiments show that our proposed method dramatically improves the speed of dynamic gesture recognition and demonstrates superior recognition results.