Low Power Gesture Sensing System Based on Target Range Using Spiking Neural Networks for Portable Devices

Over the last decade, radar-based systems have become a popular approach in intelligent human-computer interfaces due to their ability to work in any environmental condition while maintaining user privacy. One of the primary concerns in creating embedded solutions for hand gesture recognition systems is energy efficiency. To accomplished this goal, we propose an embedded HGS that utilizes spiking neural networks (SNNs) in conjunction with a 60 GHz frequency modulated continuous wave radar. The SNNs operate on event-driven principles, exhibiting sparsity in both time and space, resulting in increased energy efficiency. Compared to former state-of-the-art techniques, the system exclusively relies on range data from the target, thereby eliminating the need for the fast Fourier transform and the associated overheads. The experimental findings reveal that the proposed neuromorphic implementation is considerably smaller in size and 24.01 times faster than prior methods, while maintaining a recognition accuracy of 97%, similar to the previous approaches.