Computer Vision-Based Defect Detection and Evaluation for Automatic Inspection and Condition Rating of Concrete Bridges

Le résumé fourni par la source

Effective and accurate defect detections are of great importance for structural health monitoring (SHM) of bridges. Recent development of emerging sensing techniques and advancement of deep learning algorithms show potential in capturing bridge defects (e.g., cracks). Unfortunately, previous methods heavily rely on a high-quality imagery data set. Besides, all such methods only focus on detections of independent defects without considering the spatiotemporal distributions and correlations of numerous bridge defects. How to quickly capture spatiotemporal distributions and correlations of numerous bridge defects for evaluating health conditions of bridges is still challenging. This study has established a computer vision-based defect detection and evaluation method for automatic inspection and condition rating of concrete bridges. The proposed method utilizes the Yolov8 algorithm to first detect defects in concrete bridges and extract detailed information (e.g., dimensions and directions). Then, corresponding locations have been recorded for each individual defect. Lastly, an automatic evaluation model was developed and trained based on structural engineering knowledge for rating health conditions of bridge structures based on the identified bridge defects. Results show that the proposed method could achieve effective and accurate defect detections and condition ratings of bridges.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.