Improving road damage maintenance in South Africa using deep learning

The preservation of the road infrastructure is one of the essential factors for a safe, economical and sustainable transport system. Manually collecting data is tedious. This field is intended to benefit from the advancement of artificial intelligence technologies. Advances in deep learning enable the automatic detection of road damage from the collected road images. This work proposes to use an Indian subset of the Road Damage Dataset (RDD) 2022, which has a plethora of images of streets worldwide. The data is processed and labelled. Then, a YOLOv5s model is trained and validated. The model is evaluated against 1959 test images, and the results are tabulated and discussed. The proposed approach has the F1 results of 41% for road damage data collected from the RDD 2022 India subset. In the future, it is recommended to use the broader Global RDD 2022 data to train more robust models with higher accuracy.