Flying Watchdog: A Drone with Edge AIoT for Residential Safety and Fall Detection by Face and Posture Recognition
Kai-Jie Chang, Chung-Wei Chuang, Jun-Tung Chiu, Jen‐Yeu Chen · 2022
The high accuracy of applying deep learning on image recognition induces a variety of useful applications amid which incorporating the maneuverability and mobility of drones or UAVs for real-time monitoring can further extend the coverage of surveillance into some areas that fixed cameras cannot be installed or high resolution image cannot be obtained. Therefore, this paper proposes a system named Flying Watchdog which applying deep learning to online recognition on the images extracted from the video recorded by a drone, the flying watchdog, which can recognize a target of interest and then automatically flies toward it for further inspection. The functions of our system, like a real watchdog, consists of two major themes: home security and autonomous company. The drone with a camera on board to record the real-time surveillance video is the terminal device in an Artificial Intelligence of Things (AIoT) system. Due to the limited computing capability of a drone, through a wireless network, the video then streaming back to a mobile phone or a laptop that acting as an Edge AI computing server to execute the required procedures such as target recognition, decision, adequate alarms as well as user notifications and the drone’s maneuver control for tracking targets. By the mature deep learning model, FaceNet, friends or family members can be accurately distinguished from strangers or intruders and by the model of OpenPose, falling down of people under guard will be immediately detected. Whenever such abnormalities are found, the Edge AI computing server will immediately send notification messages to the mobile App. on corresponding users’ (masters’) mobile phone. Furthermore, the drone, the flying watchdog, can autonomously track the people of interest such as a stranger for further inspection or warning or a person under guard for detecting the possible fall. The proposed system, Flying Watchdog, is well implemented and tested and is proved to be helpful to home security and a loyal, selfless, trustworthy companion to its human masters.