Pavement marking construction quality inspection and night visibility estimation using computer vision

Pavement markings provide roadway information necessary for safe and comfortable operation. To ensure their functionality, appropriate maintenance and inspection are important. This study develops a full-scale testbed consisting of various road design parameters including marking material types, beads types, and amount of beads. Then using the field-collected images and associated retro-reflectivity (RL), Computer Vision (CV) based analysis are performed. Parameters used for examining the pavement marking construction quality are extracted to correlate with RL. In addition, a machine learning algorithm is developed to classify the RL class (from Class I to Class IV, based on RL values). Based on the CV analysis, a marking material that resulted in a deeper embedment and bead types that were prone to scatter in the test bed were revealed. Also, the overall accuracy of 82% is achieved from a transfer learning-based model, demonstrating the potential for using CV and ML algorithms for road line visibility maintenance.