← Back to Sessions

Construction of 3D Finite Element Meshes from Drone Images: A Step Towards a Non-destructive Testing Framework for Engineering Structures and Their Response to Earthquakes

Description: 

We have developed a workflow to rapidly create 3D finite element meshes, which can then be used for viscoelastic wave propagation, starting with 2D photos acquired by an inexpensive consumer-level unmanned aerial system (i.e., a drone). Using open-source and low-cost industrial computer graphics and vision software, we then invert for a 3D surface composed of triangles. This 3D surface model is enclosed and then remeshed into hexahedra for use in spectral element wave solvers. Significant advances in hardware (digital cameras, GPS, drones, and GPUs) to more flexibly capture and process higher quality images coupled with improvements in key computer vision algorithms (namely, Structure-from-Motion and Multi-View Stereo) have made it practical to reconstruct 3D surface models from thousands of images – without the need to resort to expensive equipment, esoteric calibration methods, or priors (e.g., camera settings, motion, etc.). The time required to construct accurate 3D models is often significant and so the aim of this work is to minimize that overhead.

We have tested a first iteration of this workflow on the Via Contra (Verzasca) dam in Switzerland, whose results we will present. We will also discuss key stages, challenges, and learnings as we developed the processing pipeline. Next steps include further automating the workflow and integrating these meshes with sensor data to estimate the distribution of elastic parameters. Ideally, we would like to apply this methodology to historical structures in earthquake-prone regions to image and monitor their structural health.

Session: New Directions in Environmental, Seismic Hazard and Mineral Resource Exploration Studies - I

Type: Oral

Date: 4/17/2025

Presentation Time: 02:00 PM (local time)

Presenting Author: Evan

Student Presenter: Yes

Invited Presentation: 

Poster Number:

Authors