Robust Hand Gesture Tracking and Recognition for Healthcare via Recurrent Neural Network

Abstract

In the emerging fields of computer vision, robotics, and medicine, It's difficult to recognize gestures in dynamic environment. For human-computer hand gesture tracking and recognition, symmetry is required. Recent research has presented many RGB gesture recognition methods as a result of sensor technology advancements. In this research, we propose a sustainable hand gesture tracking and recognition system that can identify and recognize RGB dynamic motions in challenging environments while maintaining accuracy. Firstly, the frames are extracted from dynamic gestures. The frames are preprocessed to reduce noise and adjust light intensity. After that, the hand is detected using the Single-shot Multi-box Detector (SSD). From the extracted hand landmarks are attained via a colour-based fast marching algorithm. The features are extracted, such as polygon mesh, neural gas, and geometric shapes. On the features, the Active Bee Colony Optimizer (ABCO) is applied to reduce dimensionality. Then Recurrent Neural Network (RNN) is applied to classify the gestures. The proposed system is tested on Sign word and Egogesture datasets, providing accuracy of 93.33% and 92.67%, respectively.