Real-Time Tunnel Light Tracking with YOLOv11 using IOU-Based Metrics and Constrained Correspondence Analysis for Onboard Vision Systems

In this paper, we propose an automatic tunnel light tracking method for onboard video processing. Based on an in-depth analysis of tunnel light movements across frames, our method covers three main aspects. Firstly, tunnel lights are detected using the YOLOv11 framework. Once a high-quality dataset is prepared, the images are annotated using the useful tool of X-AnyLabeling. The YOLOv11 model, with varient parameter settings, can be trained with this dataset using the transfer learning strategy. Secondly, the detected candidates for tunnel lights can undergo post-processing based on the principle of collinearity. To achieve this, we propose using the Random Sample Consensus (RANSAC) algorithm for line fitting to detect collinearity. This is followed by spatial sorting of the detected candidates along the fitted straight lines. Ultimately, robust tracking of tunnel lights is achieved by combining Intersection over Union (IOU) metrics with correspondence constraints analysis, between two consecutive frames. The entire process of tracking tunnel lights across frames by enforcing mandatory correspondence is described in detail. The experimental results demonstrate that this method can robustly tracking the tunnel lights under complex conditions with high accuracy, which can meet the needs of efficient acceptance inspection and daily maintenance of tunnel LED lighting system.