A Single-Photon Avalanche Diode (SPAD) camera can measure individual photons at extremely high framerates. In the example image, a sunlight simulator was placed behind a target satellite and directed into the camera sensor. Notably, the standard camera saturates and most information is lost whereas the SPAD is still able to discern the target. This project looks to explore the benefits and limitations of SPAD cameras for orbital robotics applications with a particular emphasis on spacecraft pose estimation.
This project is conducted in collaboration with the Space Robotics Lab at the University of Malaga in Spain.
Project Scope
The student will be introduced to a dataset captured at EPFL, CVLab using a spacecraft target, sunlight simulator, robotic arm and optitrack system; both a conventional RGB camera and a SPAD were used for imaging. The student will annotate the dataset and then train a neural network to perform pose estimation on the annotated images.
Tasks
- Annotation of the SPAD dataset
- Training of a simple CNN pose estimator
- Systematic evaluation of varying frame rates, exposure times and bith depths
Prerequisites
- This project is intended for a masters thesis student interested in orbital robotics and imaging.
- Python programming skills are mandatory.
- Experience training a neural network is a plus.
Contact
- Dr. Andrew Price
Postdoc at CVLab and eSpace
andrew.price@epfl.ch - Dr. David Rodriguez
Research Scientist
University of Malaga Space Robotics Lab
david.rm@uma.es