Drone-Based Real-Time Traffic Analysis: Object Tracking and Speed Measurement with YOLO, DeepSORT and Hybrid Speed Calculation

In this study, a real-time traffic analysis system was developed using drone-based object detection, tracking, and hybrid speed estimation methods. The system integrates YOLOv8 for object detection, DeepSORT for object tracking, and a hybrid approach combining GPS/Haversine and pixel-based speed estimation to provide a high-accuracy traffic monitoring solution. A drone operating at an altitude of 100 meters captures images, which are processed for object detection and tracking, while vehicle speeds are estimated using the hybrid approach. Experimental results show that object tracking accuracy remains above $\mathbf{8 5 \%}$ in low-density traffic scenarios but decreases as traffic density increases. The rise in ID switch errors under heavy traffic conditions indicates that incorporating Re-ID algorithms could improve tracking accuracy. Speed estimation results demonstrate that the GPS/Haversine method provides more stable measurements over long distances, whereas the pixel-based method offers higher accuracy in short-range scenarios. The proposed hybrid method effectively combines the strengths of both approaches, reducing overall error rates. The system was tested under various weather and lighting conditions, revealing a decline in object detection accuracy in low-light, rainy, and foggy environments. Real-time processing evaluations with Jetson Nano and an external GPU showed that the system operates at $\mathbf{3 0}$ FPS and can reach up to 50 FPS with GPU acceleration. These findings indicate that drone-based traffic monitoring systems offer significant advantages over traditional systems in terms of largescale coverage and real-time analysis.