An Ultra-low-power High-precision Dynamic Gesture Recognition Coprocessor Based On RISC-V Architecture

摘要

Contactless Dynamic gesture recognition (DGR), which has the extreme advantages of simple action, wide and strong application, faces challenges such as real-time, long-time work, low power, high complex algorithm, and more. In this article, we propose an ultra-low-power high-precision dynamic gesture recognition coprocessor based on the RISC-V architecture (UHDA). UHDA integrated on RISC-V ecological environment is configured and controlled by extended instruction set architecture (ISA). And UHDA has system feedback to adjust the background and human behavior. We adopt a low-complexity high-precision algorithm based on YCrCb and real-time online test the usual applied four gestures. The algorithm of detection gets an accuracy of 87%~100%. The final result shows that the recognition power of the UHDA with clock-gating in 28nm technology is 33.32 µw/frame. When UHDA runs at long-term standby mode, it saves more than 39.15% power than recognition power.