MACHINE LEARNING APPROACH FOR RADAR-BASED HAND GESTURE RECOGNITION USING DIFFERENT FEATURE EXTRACTION METHODS

In this study, seven different hand gestures are classified using a continuous wave radar sensor operating at 24 GHz for contactless human-machine interaction. Pre-processing is applied to the radar signals, consisting of I (in-phase) and Q (quadrature) components, to reduce noise and preserve meaningful frequency components. Subsequent to the preprocessing step, comprehensive features are extracted in the time, frequency, and statistical domains. The power spectral density (PSD) features obtained using the Welch method contribute significantly to the results. The resulting high-dimensional feature set is then subjected to classification through the utilization of conventional machine learning methodologies. The Bagged Tree algorithm, which is based on an ensemble learning approach, has been shown to outperform other traditional methods. The developed system has been demonstrated to achieve a detection rate of hand movements that exceeds 90% accuracy. This application, which is low-cost in terms of both material and processing power, demonstrates the efficacy of radar-based contactless control systems in human-machine interfaces. The findings of this study demonstrate that traditional machine learning methods continue to be a preferred approach in this context.