Research on Gesture Recognition Algorithm Based on Lightweight YOLOv4

Aiming at the problems of a large number of parameters and low detection speed of the gesture recognition model based on the YOLOv4 algorithm, a lightweight object detection algorithm n-YOLOv4 was proposed in this study. Firstly, GhostNet was used to replace the YOLOv4 backbone feature extraction network, and deep separable convolution was used to replace the $3\times 3$ conventional convolution of the non-backbone feature extraction network, which reduced the number of parameters in the model. Secondly, the CSC structure was used to replace part of the $1\times 1$ conventional convolution in the network to expand the receptive field of the network. Finally, residual edges and quintic convolution modules were added to enhance the feature extraction capability of the PANet structure. In addition, the idea of transfer learning was introduced in the training process to save training time, and the data enhancement method was used to improve the generalization of the model. The experimental results showed that when the detection accuracy was only slightly affected, the weight file size decreased by 76.72 %, and the FPS increased by 10.85 %. This method provides a new idea for the research of gesture recognition algorithms.