On the Applicability of Multimodal Neural Network Methods for Determining the Quality of the Road Surface

This paper discusses an approach to detect and classify roadway defects such as cracks and potholes based on a multimodal neural network approach. A model based on Faster R-CNN architecture has been used to detect pavement faults in images. Transformer based deep learning model SwinTransformer small was used to classify pavement damage in images. PointNet++ model was used for point cloud classification task. The multimodal pavement defect estimation method combines the predictions of three different models to improve the accuracy of the system.