부산대학교 도서관

The initial design and evaluation of FaceReminder: a real-time facial recognition app that aims to assist people with memory problems, are presented in this work. Based on a smart Vuzix Blade glass paired with an Android device, FaceReminder can show the name of a person on the see-through display of the glass when s/he is recognized. Given the limited computation and battery capacities on the smart glass, it is imperative to design FaceReminder with both performance and energy efficiency being considered. We evaluate the trade-offs between the latency of FaceReminder and power consumption of the glass with varying resolutions of the photos to be transmitted from the glass to the Android device. The evaluations also considered the size of the database used by the face recognition function. The results show that varying the resolution of the photos have the largest impact on the latency (i.e., execution times) and energy efficiency. With low resolution photos, FaceReminder can process up to three times more photos compared to the full resolution case. However, photo resolution can also affect facial recognition accuracy, which will be evaluated in our future work.