Hand Gesture Recognition using YOLO Models for Hearing and Speech Impaired People

Sign language recognition can help hearing and speech impaired persons to communicate with the rest of the world. Hand gesture recognition is widely used in the entertainment sector for cars, gaming, and other devices. In the healthcare industry, clinicians can utilise this recognition technique to handle digital images instead of touch screens or computer keyboards during medical procedures where sterilization is necessary, as well as to automatically recognise surgical gestures. For real-time, image-based hand gesture identification, recent improvements in machine learning and object detection approaches provide improved and more efficient results. We used the deep learning-based object detection models YOLOX and YOLOv5 to work on five different hand gestures for recognition in our research project. Both models were released at the same time, but YOLOX was significantly more precise and faster. For YOLOv5, we got the highest mAP score of 98%, the lowest mAP score of 89.6%, the precision score of 98.8%, and the recall score of 82.6%. Our best mAP for YOLOX was 99.55%, while our last mAP was 98.5%. In this research study, we evaluated the performance of the two models and investigated changes in the architecture that enabled YOLOX to perform so much better than YOLOv5.