Authors:
Alexander Vincent Lewi, Chihaya Morita, Shaik Ahamed, Eya Ben Slimane, Rejwan Bin Sulaiman, Ahmed Radwan
Addresses:
Department of Computing and Information Systems, Singapore Management University, 81 Victoria Street, Singapore. Department of Information and Communication Engineering, The University of Tokyo, Bunkyo-ku, Tokyo, Japan. Department of Artificial Intelligence, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Intelligent Transportation Systems, LITIS - IT Laboratory, Information Processing and Systems, University of Rouen Normandy, Mont-Saint-Aignan, Normandy, France. Department of Computer Science and Technology, Northumbria University, Newcastle upon Tyne, England, United Kingdom. McMaster Institute for Transportation and Logistics, McMaster University, Hamilton, Ontario, Canada.
Abstract:
Drones are everywhere now, and that’s created a big need for smarter, faster ways to spot and track them—whether it’s at power plants, airports, big public events, or sensitive sites. Radar and sound-based systems no longer work, especially when drones fly low. Researchers used the lightest version of a top-notch object detector, YOLOv8 Nano, to build and test a drone detection system for resource-limited PCs. Our system uses Ultralytics YOLOv8 and the Supervision annotation library. For each video frame, it draws boxes around what it sees and labels them with its confidence that it saw a drone. Researchers used ImageIO to run identified, labelled material via an H.264 video pipeline. This solves the obstinate and surprisingly prevalent OpenCV issue where MP4 movies won't play in current browsers. Researchers put everything in a simple Flask web app for user convenience. Upload a video, let the server do its AI magic, and see the labelled video in your browser—no need to install Python or deep learning tools on your machine. Researchers tested it on an NVIDIA GeForce RTX 2050 and discovered it ran at 28–35 FPS on full HD video, satisfying the real-time 25 FPS criterion. The system had 68% confidence in its detections, matching COCO-trained weights for drone, bird, plane, and kite aerial photos. Researchers identified visual detection quality acceptable for production in bright sunlight, shadows, crowded backgrounds, and various resolutions. This technology provides a solid, extendable base. With simple tweaks to tagged drone photos, it can catch the real thing better- the next step.
Keywords: Drone Detection System; Deep Learning (DL); Sound-Based Systems; Object Detector; Labeled Drone Images; Resource-Limited Computers; Sensitive Sites.
Received on: 05/10/2025, Revised on: 24/11/2025, Accepted on: 01/12/2025, Published on: 07/06/2026
DOI: 10.64091/ATICR.2026.000345
AVE Trends in Intelligent Computing Research, 2026 Vol. 1 No. 2 , Pages: 138-147