Abstract
Early monitoring road conditions and defect detection are an important step in ensuring road safety. The work presents a new road damage segmentation dataset SegmRDD. It contains 4420 images with defects of three classes "cracks", "alligator crack", "potholes" well annotated at the pixel level. The dataset is balanced and covers the roads of five countries, including Russia. Developed ensemble model based on three parallel-trained neural network models YOLOv8, U-Net, Mask R-CNN with combining results, and achieved an F1-score of 70% for all defects.
Full Text
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call