Performance evaluation of YOLOv11-based vehicle detection and tracking for urban intelligent transportation systems

Abstract

This paper proposes and evaluates an integrated vehicle detection and tracking framework based on you only look once (YOLO)v11 for intelligent transportation systems (ITS). The combination of the deep simple online and real-time tracking (DeepSORT) algorithm helps maintain vehicle identity across consecutive frames, thereby enhancing the stability of the multi-object tracking system. Additionally, the slicing-aided hyper inference (SAHI) technique is integrated to improve the detection efficiency of small vehicles in remote sensing imagery and urban surveillance video data collected in Thai Nguyen, Vietnam. The system's performance is comprehensively evaluated through several key quantitative indicators, including mean average precision (mAP), multiple objects tracking accuracy (MOTA), and identification F1-score (IDF1), across realistic urban traffic scenarios. The results show that the framework significantly improves detection accuracy, tracking consistency, and small object recognition efficiency in real-world urban traffic scenarios. This paper provides useful insights for selecting appropriate detection and tracking configurations in ITS applications.