Improving classification of road surface conditions via road area extraction and contrastive learning

TL;DR

The paper addresses automatic road surface condition classification by mitigating distractors from non-road content through a two-stage pipeline: first perform road-area segmentation to crop road regions, then classify road conditions on these crops. It introduces contrastive learning on top of a supervised classifier to stabilize embeddings and reduce intra-class variance. Empirical results on the RTK dataset show consistent improvements over strong baselines, with segmentation-driven inputs and contrastive training yielding notable gains across seven road-condition classes. The approach offers a more efficient and robust solution for road maintenance analytics relevant to ADAS, autonomous navigation, and infrastructure monitoring.

Abstract

Maintaining roads is crucial to economic growth and citizen well-being because roads are a vital means of transportation. In various countries, the inspection of road surfaces is still done manually, however, to automate it, research interest is now focused on detecting the road surface defects via the visual data. While, previous research has been focused on deep learning methods which tend to process the entire image and leads to heavy computational cost. In this study, we focus our attention on improving the classification performance while keeping the computational cost of our solution low. Instead of processing the whole image, we introduce a segmentation model to only focus the downstream classification model to the road surface in the image. Furthermore, we employ contrastive learning during model training to improve the road surface condition classification. Our experiments on the public RTK dataset demonstrate a significant improvement in our proposed method when compared to previous works.

Figures (4)

• Figure 1: Our proposed framework consists of two stages: road area extraction and classification of road surface conditions.
• Figure 2: Classification model for road surface conditions with contrastive learning to enhance performance of classification task.
• Figure 3: Ablation study with baseline is Hsieh method_raveling. '+Cont(Ours.)': training baseline classifier with our incorporated contrastive learning, '+Seg(Ours.)': using our extracted road area with segmentation for training and testing with baseline classifier.
• Figure 4: Comparison of our method with model Hsieh method_raveling in terms of ratio of confidence score of predicted road condition class.