Towards a FOSS Automatic Classification of Defects for Bridges Structural Health Monitoring

• Published in: Geomatics and Geospatial Technologies Vol. 1507; pp. 298 - 312
• Main Authors: Belcore, Elena, Di Pietra, Vincenzo, Grasso, Nives, Piras, Marco, Tondolo, Francesco, Savino, Pierclaudio, Polania, Daniel Rodriguez, Osello, Anna
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Bridge defect detection; FOSS; Machine learning; MicMac; OBIA; Photogrammetry; Random forest; Unmanned aerial vehicles (UAV)
• ISBN: 9783030944254; 3030944255
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-030-94426-1_22

Bridges are among the most important structures of any road network. During their service life, they are subject to deterioration which may reduce their safety and functionality. The detection of bridge damage is necessary for proper maintenance activities. To date, assessing the health status of the bridge and all its elements is carried out by identifying a series of data obtained from visual inspections, which allows the mapping of the deterioration situation of the work and its conservation status. There are, however, situations where visual inspection may be difficult or impossible, especially in critical areas of bridges, such as the ceiling and corners. In this contribution, the authors acquire images using a prototype drone with a low-cost camera mounted upward over the body of the drone. The proposed solution was tested on a bridge in the city of Turin (Italy). The captured data was processed via photogrammetric process using the open-source Micmac solution. Subsequently, a procedure was developed with FOSS tools for the segmentation of the orthophoto of the intrados of the bridge and the automatic classification of some defects found on the analyzed structure. The paper describes the adopted approach showing the effectiveness of the proposed methodology.