Autonomous Aerial Surveillance for Intelligent Pothole Detection and Alert System

Manual pothole inspection is slow and resource-intensive, leading to delayed repairs and increased road hazards. This paper presents an autonomous aerial surveillance system that integrates a custom UAV, a YOLOv8 detection framework, and a real-time mobile alert application. Trained on 826 annotated aerial images, the model achieved a precision of 0.84 and mAP@0.5 of 0.89 with filters based on bounding-box area <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(&gt; 8000\text{px}^{2})$</tex> and confidence <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(&gt; 0.7)$</tex>. The Android application provides multimodal alerts (voice, visual, SMS) within a 100 m radius, with average response latency under one second. The 3.5 kg quadrotor, equipped with a downward-facing 4K camera and RTK-GPS, supports 53-minute flights covering 22.4 km per mission. This end-to-end system enables scalable pothole detection, geotagged mapping, and proactive driver warnings, contributing to safer and more efficient urban transportation infrastructure.