Metodologia para monitoramento de estruturas utilizando descritores visuais

Abstract

Monitoring large-scale infrastructures, such as dams or those built on sloped foundations, is essential for ensuring their structural integrity and safe use. The conventional monitoring process utilizes several sensors spread throughout the structure. This limits the detection and calculation of deformities to areas where the sensors are positioned. Photogrammetry techniques combat this limitation by monitoring entire areas at low-cost, using only a camera. In this context, the Multiscale Model to Model Cloud Comparison (M3C2) technique is considered state-of-the-art for structural deformation calculation in point clouds. However, M3C2 incorrectly assumes that all structural deformities occur normal to the monitored surface, which can lead to inaccurate deformation estimates. Consequently, this work proposes a novel methodology for calculating deformations in infrastructure, entitled Visual Correspondence and Point Reprojection Distance. This method uses visual descriptors to find corresponding points among point clouds captured at different times. Validation of the proposed technique was performed in three controlled experiments using artificially created deformations, and one test conducted in a rock slope infrastructure. The results show that the proposed method is superior to M3C2, especially when deformations are not normal to the structure’s surface.