Tiny Object Detection in Aerial Traffic Surveillance using YOLOv10-Nano

Detecting tiny objects in aerial traffic surveillance remains a significant challenge due to minimal object scale, frequent occlusions, and dense environments. This study investigates the performance of the lightweight YOLOv10-Nano (YOLOv10n) model for tiny object detection using the VisDrone dataset—a benchmark recognized for its real-world complexity. The research evaluates the model’s accuracy, processing latency, and edge deployment viability, particularly on devices such as the NVIDIA Jetson Nano. To improve detection of small objects, enhancements including the ERAC module and tailored training techniques were applied. Experimental outcomes demonstrate that the modified YOLOv10n surpasses models like YOLOv5n and SSD in both detection precision and real-time performance. The findings affirm YOLOv10n's potential in enabling efficient, real-time aerial surveillance and present practical strategies for deploying such models on resource-limited platforms.