Monocam Drone Swarm Guard: A Real-Time Collision Detection and Avoidance System

Abstract

Another technology which is becoming more popular in our skies is the unmanned aerial vehicles (UAVs or drones) particularly when they are operated in swarms. With the increased use of drones’ concern over mid-air collisions has also become a natural issue particularly where visibility is low or when the drones are operating within a very close radius. This paper presents Monocam Drone Swarm Guard- a system that enables drones to navigate a shared airspace with one camera safely. Rather than implementing an intricate sensing system we decided to operate the system on a normal laptop with live video and discovered that it was able to follow through with drones even under vegetation and motion blur conditions. Our system can detect potential collisions before they happen by applying deep learning (YOLOv8) to detect and using a combination of a Kalman and Extended Kalman Filters to track and predict drone paths. Based on our experiment the findings indicate that the approach is effective in realtime tracking and ensuring safety of drone swarms when their operation occurs in adverse and realistic environments.