Prediction and detection of potholes in urban roads: Machine learning and deep learning based image segmentation approaches
Sang-Yum Lee, Tri Ho Minh Le, Yeongmin Kim
Abstract
The number of potholes in the world has rapidly increased due to the growth of vehicles, temperature changes, and the concentration of the population. Potholes cause danger in driving and reduce passengers' comfort. Therefore, an accurate prediction of number of potholes provides timely maintenance and rehabilitation, and also it enhances safety for drivers. This study aims to improve the accuracy of number of potholes prediction model by considering independent variables such as minimum temperature, relative humidity, precipitation, and traffic volume. The model was established by conducting variable analysis. Various machine learning methods were then employed to develop an optimal model that provides the highest accuracy in predicting pothole occurrence. The study also suggests a computer vision-based system for spotting potholes based on the image segmentation method, followed by calculating the damage ratio. The results confirm that the proposed models have the potential in predicting and detecting pothole occurrence.