Autonomous Target Detection in Drone-Based Warfare Systems Using OpenCV and Deep Learning Object Detection Frameworks
Autonomous Target Detection in Drone-Based Warfare Systems Using OpenCV and Deep Learning Object Detection Frameworks
Authors:
Dr. Shagufta Mohammad Sayeed Sheikh, Aum Patil, Prithviraj Chavan, Vipul Barmukh
Abstract
Unmanned Aerial Vehicles (UAVs), commonly known as drones, have fundamentally transformed the landscape of modern warfare through the integration of real-time computer vision and artificial intelligence. This paper presents a critical analysis and technical framework for autonomous target detection in drone-based warfare systems, leveraging OpenCV, YOLO (You Only Look Once), and deep convolutional neural networks (CNNs). We investigate the implementation of real-time object detection pipelines aboard edge-deployed platforms such as NVIDIA Jetson, examining performance trade-offs between detection accuracy (mAP) and inference latency under battlefield constraints including low-altitude imaging, thermal signatures, and occlusion. Our system achieves a mean Average Precision (mAP@0.5) of 87.3% on the VisDrone2023 benchmark dataset at 28.6 FPS on embedded hardware. We additionally propose a lightweight Temporal Consistency Filter (TCF) that suppresses spurious detections across consecutive frames, yielding a further 2.1% mAP improvement with negligible latency overhead. Comprehensive ablation studies are conducted to isolate the contribution of each pipeline component—preprocessing, backbone depth, quantization strategy, and post-processing—providing practitioners with actionable design guidance. We critically evaluate ethical concerns, failure modes, adversarial vulnerabilities, and International Humanitarian Law (IHL) compliance challenges associated with autonomous lethal systems, offering a balanced assessment of the current state of the art and future research directions.
Index Terms
UAV, drone warfare, OpenCV, YOLO, object detection, autonomous systems, computer vision, edge AI, VisDrone, convolutional neural networks, IHL compliance, temporal consistency filtering, model quantization, adversarial robustness.