Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Optical Flow Based Velocity Estimation for a Tunnel Fire Backlayering Scenario
This project focuses on motion extraction from synthetic BOS images for a tunnel fire backlayering scenario. The student’s task will include the assessment, optimization and probabilistic formulation (e.g. using Bayes theorem) of optical flow techniques for flow velocity estimation.
The background oriented schlieren (BOS) technique is sensitive to temperature variations in fluids and provides projective 2D data of coherent flow structures. These 2D flow signatures can serve as traceable structures to estimate flow velocities using for instance optical flow techniques. The tunnel fire backlayering scenario represents an interesting test case for BOS velocimetry applications due the occurrence of counter-moving flow velocities.
Optical flow is a well-established method for motion detection. However, there is an inherent uncertainty in the underlying brightness constancy assumption due to the turbulent behavior of the flow and the integrative nature of the (speckle) BOS technique. This project will mainly focus (but is not necessarily limited) on the optimization and probabilistic extension of optical flow estimates for the backlayering scenario using synthetic data. As an example, Bayesian approaches can be used to incorporate prior knowledge (e.g. from CFD data) into the estimation problem and to regularize the inherently ill-posed nature of the optical flow problem.
This project does not require physical presence at ETH.
The background oriented schlieren (BOS) technique is sensitive to temperature variations in fluids and provides projective 2D data of coherent flow structures. These 2D flow signatures can serve as traceable structures to estimate flow velocities using for instance optical flow techniques. The tunnel fire backlayering scenario represents an interesting test case for BOS velocimetry applications due the occurrence of counter-moving flow velocities.
Optical flow is a well-established method for motion detection. However, there is an inherent uncertainty in the underlying brightness constancy assumption due to the turbulent behavior of the flow and the integrative nature of the (speckle) BOS technique. This project will mainly focus (but is not necessarily limited) on the optimization and probabilistic extension of optical flow estimates for the backlayering scenario using synthetic data. As an example, Bayesian approaches can be used to incorporate prior knowledge (e.g. from CFD data) into the estimation problem and to regularize the inherently ill-posed nature of the optical flow problem. This project does not require physical presence at ETH.
- Familiarize with optical flow techniques and tunnel fire backlayering scenario.
- Optimization of Lucas-Kanade and Horn-Schunck algorithms for the backlayering problem.
- Review and implementation of probabilistic optical flow techniques.
- Documentation and presentation of the results.
Depending on the student’s background, the project may also include supervised or unsupervised learning for optical flow estimation.
- Familiarize with optical flow techniques and tunnel fire backlayering scenario. - Optimization of Lucas-Kanade and Horn-Schunck algorithms for the backlayering problem. - Review and implementation of probabilistic optical flow techniques. - Documentation and presentation of the results.
Depending on the student’s background, the project may also include supervised or unsupervised learning for optical flow estimation.
Interested candidates are invited to send an email with a transcript of record to buehlmann(ad)ifd.mavt.ethz.ch.
Interested candidates are invited to send an email with a transcript of record to buehlmann(ad)ifd.mavt.ethz.ch.