<![CDATA[This study develops a drone-based multi-class traffic counting system using fine-tuned YOLO11 integrated with the SORT tracking algorithm on a one-way urban corridor. Aerial data was captured at an altitude of 25 meters using a DJI Mavic 3 drone from a nadir perspective to minimize intermodal occlusion. System performance was validated against manual ground truth using Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE) metrics. The system achieved an overall MAPE of 18% and an RMSE of 45.07. Notably, the car class demonstrated perfect accuracy (0% error), confirming that these automated counts are suitable for direct application in fundamental traffic engineering metrics. Conversely, significant overcounting occurred in the motorcycle class (+34.9%), primarily attributed to Non-Maximum Suppression (NMS) failures under conditions of dense spatial proximity. In civil engineering practice, utilizing uncalibrated automated counts risks overestimating the Volume-to-Capacity (V/C) ratio, leading to a false degradation of the reported Level of Service (LOS). Consequently, a specific calibration factor of -26% for motorcycle counts is essential to ensure data validity for high-precision infrastructure design and signal timing optimization. Keywords: Drone, Traffic Counting, NMS, SORT, YOLO]]>
Copyrights © 2026
