Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Camera to GPS calibration for trains using maps
The ability to detect obstacles on the vehicle's path is crucial for developing autonomous vehicles, such as trains or cars. Detecting objects early enough and, therefore, at a long range is crucial, especially for heavy vehicles like trains. It is helpful to know where the train tracks and other infrastructure elements are for vision-based systems. Rather than using machine learning approaches, which require extensive labeled training data and long inference times, one could reproject a known railway map into the camera view. Estimating the camera's exact position and rotation relative to the vehicle, and achieving a good alignment, is the challenge here.
In this project, we aim to create a camera map overlay, which various downstream vision applications require. Typically, a GPS-to-camera calibration uses calibration aids and a particular procedure. In the field or with existing datasets, those are typically unavailable. We aim to develop a calibration procedure based on visual cues and available map information. The challenges lie in integrating infrastructure measurement into an optimization framework and temporally consistent sensor fusion.
In this project, we aim to create a camera map overlay, which various downstream vision applications require. Typically, a GPS-to-camera calibration uses calibration aids and a particular procedure. In the field or with existing datasets, those are typically unavailable. We aim to develop a calibration procedure based on visual cues and available map information. The challenges lie in integrating infrastructure measurement into an optimization framework and temporally consistent sensor fusion.
- Review of literature on extrinsic calibration.
- Familiarization with existing code and work in this field.
- Development of a continuous calibration and reprojection pipeline.
- Testing and evaluation using recorded or simulated datasets.
- Review of literature on extrinsic calibration. - Familiarization with existing code and work in this field. - Development of a continuous calibration and reprojection pipeline. - Testing and evaluation using recorded or simulated datasets.
- Highly motivated and independent student.
- Interest in SLAM and sensor fusion.
- Programming skills in Python or C++.
- Experience with computer vision frameworks is a plus.
- Enrolled at ETH Zurich.
- Highly motivated and independent student. - Interest in SLAM and sensor fusion. - Programming skills in Python or C++. - Experience with computer vision frameworks is a plus. - Enrolled at ETH Zurich.
If you are interested, please send your CV, Transcript of Records and a short paragraph explaining your motivation for the project to: cornelius.voneinem@mavt.ethz.ch, david.hug@mavt.ethz.ch, rik.baehnemann@mavt.ethz.ch
If you are interested, please send your CV, Transcript of Records and a short paragraph explaining your motivation for the project to: cornelius.voneinem@mavt.ethz.ch, david.hug@mavt.ethz.ch, rik.baehnemann@mavt.ethz.ch