Hand Gesture Recognition Using EfficientNetB5: A Robust Approach for Real-Time Human-Computer Interaction

Abstract

Human-computer interaction depends on hand gesture recognition, which has a major influence on virtual reality, robotics, and assistive technology among other domains. This work investigates the EfficientNetB5 convolutional neural network model's application to precisely categorize eight unique hand gestures: closed fist, finger circle, finger symbols, multi-finger bend, open palm, semi-open fist, semi-open palm, single finger bend. Drawn from Kaggle, the dataset consists of 1,841 photos guaranteeing a varied depiction of motions to improve model training. Data preparation, model architecture building, k-fold cross-valuation-based training, and performance evaluation comprise the suggested approach. Our findings show a good general accuracy of 90%, with perfect categorization achieved from particular gestures. Particularly in identifying more complicated motions, a classification report and confusion matrix analysis exposes the strengths and opportunities for development of the model. The results highlight the possibilities of EfficientNetB5 in real-time hand gesture detection systems, hence advancing understandable human-computer interactions.