A gesture recognition algorithm based on deep learning

Sign language is an important tool for hearingimpaired people to express their thoughts and communicate, for the existing sign language recognition algorithms existing low detection accuracy, high leakage rate and poor resistance to environmental interference, this paper is based on the YOLOv5 algorithm, proposes an improved method of sign language recognition, the YOLOv5 algorithm is improved as follows: using the RFCAConv balanced sensory wild-attention convolutional operation to reconfigure the C3 module, the C3_RFCAConv module is proposed, followed by applying the C3-DCN module in order to capture the finger changes of sign language movement and reduce the background interference, and finally, BiFPN is designed by integrating BiFPN with the EfficientChannelAttention (ECA) attention mechanism. The experimental results show that the improved sign language recognition algorithm has increased the mAP@50 metric by $10.9 \%$ compared to the original YOLOv5 model and also achieves more accurate detection results.