Wearable Gyroscope-Based Motion Tracking and Hand Gesture Recognition Using ANOVA for Feature Discrimination

Abstract

Hand gesture recognition has been identified as an important field of research in the healthcare sector, specifically for applications in rehabilitation, assistive communication, and human computer interaction. Upper limb motion tracking provides useful information when determining the recovery of movement in patients suffering from conditions such as stroke, Parkinson’s disease, spinal cord injuries, and others. Existing methods utilize camera-based systems or EMG sensors; however, their effectiveness may be hindered by various factors such as lighting conditions, privacy issues, electrode positioning, and computational difficulties. In this paper, a wearable hand gesture recognition system employing an MPU6500 gyroscope to detect tri-axial angular velocity corresponding to five different pre-defined hand gestures: Center, Up, Down, Tilt Left, and Tilt Right. The detected signals are then filtered through a low-pass filter and normalized before statistical features are obtained in the form of mean, variance, root mean square (RMS), and signal magnitude vector (SMV). Statistical significance of the features is assessed through one-way ANOVA while the Artificial Neural Network (ANN) model is used for classifying the gestures. The proposed device was tested on signals obtained from 19 subjects aged between 14 and 16 years. Experimental results have shown that the classification accuracy was 94.2%, while the precision was 93.6%, the recall was 92.8%, and the F1-Score was 93.1%. It can be concluded from the above results that the system employed is a portable system for gesture recognition, thus making it ideal for applications in rehabilitation monitoring, assistive communication, and wearables.