Battery-Less Face Recognition at the Extreme Edge

Machine learning-based face recognition systems are commonly used in mobile platforms to assist the camera systems, unlock the device, or analyze the facial expressions. The computational complexity of the underlying algorithms as well as the power consumption of the entire imaging and processing system largely limit the deployment to powerful mobile processing systems with large rechargeable batteries. However, these computer vision capabilities would also be useful in miniaturized low power applications with stringent battery-size limitations. We assess the feasibility of such a computer vision edge processing system on a battery-less credit card-sized demonstrator using an ultra-low power image sensor and a machine learning system-on-chip, achieving self-sustainable operation using solar energy harvesting with a small on-board solar cell. The tested system enables continuous 1 frame-per-second battery-less imaging and face recognition in indoor lighting conditions.