English title

Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:

---

Automatic detection of road damage, speeds up the process of their maintenance and prevents many traffic accidents. In this research paper, ten road damage detectors were developed in the RDD2020 dataset to check the performance of road damage detectors using YOLOv5. By simulating environmental conditions such as intense sunlight and the shadow on the road surface it was determined that road damage detectors perform poorly in these conditions. Further, by using traditional data augmentation techniques, the robustness of road damage detectors was improved in different environmental conditions. Using these techniques, a validation dataset was created to check the performance of detectors in different environmental conditions in the RDD2020 dataset. In order to solve the problem of data scarcity in this dataset, the bounding box augmentation technique and modified Poisson blending were combined. In this method, the number of instances of the reference dataset was increased in pothole and horizontal crack class. The results of this research show that training detectors with new data has improved their performance in F1-Score and mAP by 33% and 50%, respectively.

Keywords: Object Detector, Road Damage, Data Augmentation

     

Add your comments about this article : Your username or Email:

---

---