Modern video surveillance applications pose strict constraints on implemented algorithms in order to obtain low complexity, low cost, and low power consumption. Motion detection algorithms enable good power efficiency since the system may be woken up only when motion is detected. In this paper an algorithm with a high degree of reliability against noise and illumination changes is proposed and its implementation on an innovative embedded platform presented. The main goal is to demonstrate the efficiency of this algorithm using only off the shelf components instead of a dedicated structures. To conduct fair measurements we design a complete prototype able to capture video streams from an USB camera, detect the motion, compress video storing it appropriately and share it through an embedded web server.