For autonomous ground robots it is important to have a continous and reliable estimate of the robots’s movement in the local frame of reference. Visual inertial odometry (VIO) uses a camera working in the visual spectrum in combination with an inertial measurement units (IMU) to provide such an estimate. However, VIO setups are highly dependent on the existing light conditions, and for example darkness or fog will severly limit their performance.
Replacing the visual spectrum camera with a thermal imaging system could potentially mitigate these issues. This approach has previously been explored, for example by [], which uses a thermal imager mounted on a car, and [] which uses a deep learning framework for indoor navigation. There is however a lack on methods focusing on small ground vehicles, where different assumptions have to be made about the computational capabilities, the enironment and the agents movement. We are also mainly interested in reliable odometry, not in large scale or long term positioning, i.e. SLAM.
We want a student that is interested in helping us with developing and evaluating a reliable method for thermal inertial odometry. The method should be suitable for small autonomous ground vehicles with limited computational capabilities. Focus should be on reliability in the short term, local frame.
Here are som suggested areas of focus:
- Development/Evaluation of computationaly efficient feature extraction algorithms for thermal images.
- Evaluating optical flow for thermal images.
- Development of stereoscopic themal imaging.
- Exploring practical consequences of using thermal cameras instead of RGB for small ground vehicle odometry.
We are of course also open to suggestions.
- An interest in (classical) Computer Vision and perception.
- Some prior knowledge about localization of autonous system.
- Programming experience, ideally in C++.
- Ideally you have prior experience with embeded systems, or hands on experience with autonomous systems and ROS.
- It would be a bonus if you have a broader interest in robotics and mechatronics as well.
- Swedish citizenship is a requirement.
A mix between working from home and at our office in central Stockholm.
We offer compensation at end of project.
ACNR Cyber Technology is a young tech company that works with safety solutions and technical challenges primarily for swedish companies and organisations. Our main office is situated in central Stockholm. If you have any questions, you are welcome to contact Henrik (Henrik@acnr.se) or read more about us at www.acnr.se.
P. V. K. Borges and S. Vidas, "Practical Infrared Visual Odometry, in IEEE Transactions on Intelligent Transportation Systems, vol. 17, no. 8, pp. 2205-2213, Aug. 2016, doi: 10.1109/TITS.2016.2515625.
Saputra, M.R., Trigoni, N., de Gusmao, P.P., Lu, C.X., Almalioglu, Y., Rosa, S., Chen, C., Wahlström, J., Wang, W., & Markham, A. (2020). DeepTIO: A Deep Thermal-Inertial Odometry With Visual Hallucination. IEEE Robotics and Automation Letters, 5, 1672-1679.