Drone platform for safety in testing

Abstract

This thesis addresses the need for improved safety surveillance in autonomous vehicle testing by developing a drone platform for multiple drones. Extending single drone systems, this work focuses on coordinating multiple drones to provide wider area coverage at testing facilities. The backend of the developed system calculates drone positions based on test trajectories from ATOS, an open-source platform for automated vehicle testing, ensuring required sensor overlap. Real-time video is streamed from DJI drones via an Android app using Web Real Time Communication to a backend system. The backend processes the streams, performs image stitching to create a panoramic view, and applies object detection with estimated GPS coordinates for bikes, cars, and pedestrians. A web-based frontend allows users to monitor individual and merged video feeds, view detected objects, and control the drones. The system architecture is containerized using Docker. Experimental validations demonstrate the platform’s ability to control drones, stream video in realtime, and provide extended surveillance with object detection. Key challenges and future improvements, including addressing hardware dependencies and automating configuration, are identified. The project represents a collaborative effort between Chalmers University of Technology and Pennsylvania State University.