Conference paper Open Access
Zou,Yu; Gottardi, Massimo; Lecca, Michela
This paper reports on a low-power vision sensor embedding a custom algorithm for event detection. Anomalous or suspicious motions occurring in the scene are isolated from the background by continuously monitoring the time variation of the pixel intensity with respect to two thresholds that are dynamically updated. Pixels whose intensity is out of the range defined by these thresholds are considered as a part of a possibly anomalous activity and they are called hot pixels. The sensor has been fabricated in a 110nm CMOS technology. It delivers gray-scale images in QVGA format and related hot-pixel bitmaps at 15 fps with a power consumption of 1.6mW.