Tiny Overseer - A System for Autonomous Low-Altitude Missions

Abstract

This thesis will demonstrate a complete system based on commercial quadcopter drones with the ability to perform autonomous unmanned missions. The system is based on state of the art object detection networks with an emphasis on SSD. The goal of the system is to perform surveillance and inspection over areas of interest. During missions, the drone is streaming video and other sensor data in real-time to an external server that performs inference using neural networks and predicts the geolocation of detected targets. We attain geolocation predictions from altitudes above 25 meters with errors in the range of 1 - 7 meters. The quadcopter applied during the research is a DJI Mavick Pro.