Gesture-based Human-in-the-Loop Interaction with Fully-Adaptive Radar

Conventional radar systems transmit a fixed waveform based on desired performance metrics or experience. Recently, cognitive or fully-adaptive radar has been proposed to overcome performance degradation incurred due to environment dynamics. This work is the first to consider adaptation of the transmit signal based on interaction with a human user in the context of gesture-based control of devices or smart environments using radar. In particular, the proposed human-centered fully-adaptive radar (HC-FAR) system selects one of three different transmit waveforms based on human presence and the interactive task required (device triggering or command recognition). Moreover, the proposed system adaptively selects subsequent processing and RF data representation used at the input to a multi-input deep neural network based on the transmit waveform selected. Our results show that, HC-FAR ensures more efficient use of computational resources, while enhancing real-time gesture recognition performance in natural, real-world settings.