This paper presents a low-power programmable dynamic gesture recognition chip based on a RISC-V processor using RGB images. The chip uses two gesture recognition algorithms to recognize the most commonly used dynamic gestures (up, down, left, and right) with any shape of a hand at the distance of 10 cm $\sim$ 90 cm and speed of 10 cm/s $\sim$ 150 cm/s. The chip is fabricated in a 28 nm process and occupies an area of $640\mathrm{\mu }\mathrm{m}$ x $640\mathrm{\mu }\mathrm{m}$. The accuracy reaches 93% $\sim$ 99% at complex backgrounds with 320x240 RGB image. It operates either in the recognition mode or in the standby mode that can be automatically switched to the other. The power consumption of the recognition mode is 397 $\mu$W at 0.584 V supply voltage, 25 MHz clock, and 30 FPS. And the power consumption of the standby mode at room temperature is 78.3 $\mu$W.

本文提出了一种基于 RISC-V 处理器的低功耗可编程动态手势识别芯片，使用 RGB 图像。芯片采用两种手势识别算法，识别最常用的动态手势（上、下、左、右），距离10厘米处任意形状的手$～$ 90 厘米和 10 厘米/秒的速度 $～$150 厘米/秒。该芯片采用28nm工艺制造，占地面积为$640\mathrm{\mu }\mathrm{米}$ X $640\mathrm{\mu }\mathrm{米}$. 准确率达到93%$～$99% 在 320x240 RGB 图像的复杂背景下。它在识别模式或待机模式下运行，可以自动切换到另一种模式。识别模式功耗397 $\mu$W 在 0.584 V 电源电压、25 MHz 时钟和 30 FPS 下。常温待机功耗为78.3 $\mu$W.