Evaluating Road Crack Segmentation Performance in Participatory Sensing: An Exploration of Alternative Metrics

In participatory sensing, road cracks can be monitored utilizing the dashboard cameras (dashcams) of regular vehicles. While the predicted segmentation masks in dashcam images may seem satisfactory for assessing road crack ratios to the human eye, the intersection over union scores can be notably diminished due to subtle variations in crack thickness. To address this issue, this study introduces a new metric—crack topology score—that focuses on crack structures without being overly sensitive to crack width, utilizing a buffer approach and a dedicated algorithm. Experiments confirmed the metric’s effectiveness compared to existing metrics for segmentation results, with high scores for accurate crack segmentation and topology representation. This metric will enhance participatory sensing technology for road infrastructure monitoring, as accurate crack topology assists in determining crack ratios essential for road maintenance and repair decisions.