A centroid-based algorithm for measuring and tracking vehicle speed from a monocular camera using the YOLOv8 object detector

Abstract

Accurate real-time vehicle speed measurement is crucial for enhancing road safety and advance intelligent transportation systems (ITS). This paper proposes a centroid-based tracking algorithm that integrates YOLOv8, a state-of-the-art object detector, with DeepSORT for robust multi-object tracking. By leveraging YOLOv8’s anchor-free detection and DeepSORT’s appearance-based association, the proposed method effectively mitigates occlusions and minimizes identity switches. Evaluations on the VS13 benchmark dataset reveal a 2–5% improvement in measurement accuracy as compared to existing solutions, while maintaining real-time performance at 30 FPS. The method demonstrates consistent reliability across different vehicle models, speeds, and lighting conditions, underscoring its adaptability to real-world traffic scenarios. Moreover, larger bounding boxes enhance tracking stability, reducing false detections. Overall, the approach’s low computational overhead and high accuracy position it as a practical solution for ITS applications in constrained environments.