Drone-Based Real-Time Traffic Analysis with YOLOv8 and ByteTrack

This study presents an integrated system for realtime analysis of traffic imagery obtained from drones. The proposed method leverages YOLOv8 and ByteTrack algorithms to perform road segmentation, vehicle detection, tracking, and counting with high accuracy and low latency. The system identifies dynamic regions of interest (ROI) by pixel-level road segmentation from drone imagery, optimizing computational load by analyzing only these areas. Training utilized datasets such as COCO, KITTI, and BDD100K, alongside custom drone imagery, enhanced through transfer learning and optimization techniques to improve model performance. Tested across diverse traffic conditions (urban roads, highways, intersections) and environmental variables (lighting, weather), the system achieved over 85% accuracy and speeds exceeding 30 FPS, demonstrating its realtime traffic monitoring capability. The ByteTrack algorithm ensured vehicle identity consistency during tracking, preventing duplicate counts. The results indicate that the system offers an effective solution for smart city applications, traffic management, and data analytics.