Sign language letters recognition model based on improved YOLOv5

To solve the problems of low recognition accuracy and poor robustness of existing sign language letters recognition models in scenes such as complex background interference and overlapping hands, in this paper, a YOLOv5-SLL sign language letters recognition model is proposed. Firstly, a Sign Language Letters (SLL) dataset contains a total of 3373 annotated sign language letters images is established. Secondly, the Convolutional Block Attention Module (CBAM) is introduced into the basis of the YOLOv5 network structure, so that the network can focus more on the extraction of hand features and reduces the interference of background noise. Finally, the redundant bounding boxes are optimized by using the soft-Non Maximum Suppression (soft-NMS) algorithm, which alleviates the problem of missed detection that easily occurs when the hands overlap. The experiments are conducted on the SLL dataset, and the results show that the proposed YOLOv5s-SLL model improves the mean Average Precision (mAP) by 2.03% compared with the YOLOv5 model, under the recognition speed is unchanged. And compared with other advanced algorithms such as SSD, YOLOv3, etc., the YOLOv5-SLL model still has a higher recognition accuracy and robustness.