Vison-based 3D vibration measurement of bridges with multiple cameras/drones
Problem statement
Monitoring vibrations of infrastructure such as bridges is a popular direction in the field of Structural Health Monitoring (SHM). Traditional methods typically rely on contact-based sensors for measurements. However, with advancements in computer vision and deep learning technologies, utilizing non-contact sensors such as cameras and drones has emerged as a highly promising research direction, offering significant cost savings.
Most existing work focuses solely on planar vibrations of bridges, but vibrations in the third dimension cannot be ignored. To accurately measure the 3D vibrations of bridges, two key challenges need to be addressed: camera pose estimation and object tracking. Additionally, it may be necessary to create a camera pose estimation dataset and a 3D vibration dataset specifically tailored for bridge models, which would provide a substantial contribution to the relevant research community.
Task:
Given the numerous steps involved, the following outlines the potential tasks, which can be discussed and finalized to determine the specific ones to undertake.
1. Collect data (videos) of bridge model using fixed cameras, do camera calibration (Ground Truth).
2. Compare the effectiveness of SOTA camera pose estimation methods under sparse viewpoints, analysing their strengths and weaknesses.
3. Attempt to improve the performance of methods above for camera pose estimation in sparse viewpoint scenarios.
4. The optimal algorithm found above can be used to estimate camera poses from images taken by the drone. Combing object tracking, we can calculate multiple vibration results under different condition. Then we can compare them and do analysis.
You will get:
1. Profound experience in related fields, such as deep learning models, and 3D reconstruction.
2. A publication at top-tier AI venues if the work is qualified.
Work:
40% Theory, 40% programming, 20% Writing.
Contact:
Qingyu Xian (q.xian@utwente.nl)