Face Processing on Low-Power Devices

Abstract

The research on embedded vision-based techniques is considered nowadays as one of the most interesting matters of computer vision. In this work we address the scenario in which a real-time face processing system is needed to monitor people walking through some locations. Some face detection (e.g., Viola-Jones face detector) and face recognition (e.g., eigenfaces) approaches have reached a certain level of maturity, so we focused on the development of such techniques on embedded systems taking into account both hardware and software constraints. Our goal is to detect the presence of some known individuals inside some sensitive areas producing a compact description of the observed people. Captured data can then be used for online or offline behavioural analysis. In order to obtain system scalability and efficient resource allocation, Wireless Multimedia Sensor Networks (WMSNs) are used. Each node is based on the low-power Imote2 platform extended with a multimedia board that integrates a VGA camera and a PIR motion sensor. Tests have been performed on a WMSN prototype deployed for the monitoring of a laboratory at the University of Palermo and experimental results are very encouraging.