Retrainable Hand Gesture Recognition Model Driven by Edge AI

Based on computer vision and machine learning, accurate hand behavior and gesture recognition are crucial, especially for missing or incomplete hand models. Finger joint point prediction technology is of great value for applications such as virtual reality and human posture recognition. However, hand joint point prediction models, such as Google’s MediaPipe, perform well in full-finger models but are ineffective in missing-finger situations. Missing fingers are familiar in clinical medicine, rehabilitation facilities, and unique scenarios such as hand injuries or congenital hand deformities. Accurate prediction of hand joints is critical for hand rehabilitation, prosthetic design, and human motion analysis and inspires the development of related technologies. Since traditional prediction models rely on complete hand data, and missing fingers can make the hand structure irregular, the complexity of prediction increases. This study proposes an improved prediction model focusing on hand joint point prediction in the case of missing fingers, hoping to improve prediction accuracy and provide valuable reference for future applications and research. Among the research results, we successfully integrated the new training image model with MediaPipe, significantly improving missing fingers’ prediction accuracy.