Automatic arm gesture recognition using information gleaned from maximum instantaneous doppler frequencies

Radar has been recently employed for automatic hand gesture recognition for touchless interactive intelligent devices. Compared to hand gesture, arm gesture recognition can be more suitable for contact-less man-machine interaction with longer range separation. The larger radar cross-section of the arms, vis-a-vis hands, permits more remote interactive positions in an indoor setting. Further, the ability of using hand gestures for device control can sometimes be limited by cognitive impairments such the Parkinson disease. In this case, arm motions can be more robust to strong hand tremor and shaking. In this paper, we discriminate between dynamic arm motions using a Doppler radar sensor. The method considered is motivated by the clear contiguity of the arm signal back-scattering in the time-frequency domain. We use information gleaned from the arm micro-Doppler (MD) signature as the sole features and proceed to classify arm motions using the Nearest Neighbor (NN) classifier. In particular, we analyze the role of the maximum instantaneous Doppler frequencies and their distribution on classification performance. The proposed classification method favorably compares with other methods based on principal component analysis (PCA) and convolutional neural networks (CNN).