Visual Tracking of mini-UAVs using Modified YOLOv5 and Improved DeepSORT Algorithms

Over the past few years, unmanned aerial vehicles (UAVs) or drones have been deployed across the military services. At the same time, the civilian market has seen an exponential growth of essentially smaller systems (mini-UAVs) intended for public use. Recently, the malicious use of mini-UAVs has begun to emerge among terrorists, criminals and smugglers. The probability and frequency of the attacks of these devices, which have become more technologically advanced and cheaper, are both high and their impact can be very dangerous with devastating effects. Therefore, the need for counter-measure solution is highly required. For this purpose, the aim of this work is to propose a solution to early warning detect and track rogue mini-UAVs in restricted area. To meet this challenge, an online multiple object tracking (MOT) strategy is proposed based on YOLOv5 and the DeepSORT tracker. We adapt the former to the context of detecting small objects (ie.e. drones) by modifying the architecture. We enhance the discriminative power of DeppSORT tracker based on deep feature-based appearance. We combine detection and tracking that allows us the capability to handle small object localization. The experimental results show that the proposed multi-object tracking system is able to track mini-UAVs in real time.