Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Deep learning enhanced vision-based localization under environmental and seasonal changes
Vision-based localization in satellite imagery under seasonal and environmental changes onboard a UAV
Keywords: Computer Vision, Deep Learning, Localization, UAV
ASL has been working on fixed-wing UAVs since 2007, more recently shifting the research focus from solar powered UAVs to autonomous planes that can also operate beyond visual line-of-sight to the (safety) pilot. For example in the case of a GPS outage or failure, it is important that the UAV is able to continue its mission.
Currently, ASL is able to map an area, generate a globally consistent map and localize within the map if the environmental and illumination changes are neglectable. Reliably aligning the live image stream from the drone to satellite imagery, while relatively easy for humans, poses still a challenging task for computer vision algorithms. In this context, the project will build up on recent advances in the field of deep learning, in particular in learning landmarks that are stable over time. Furthermore, ASL has access to high precision satellite data from Switzerland (covering up to 10 years) which could be used as training data.
Related Literature
- Goforth et al., “Aligning Across Large Gaps in Time”, 2018
- Wan et al., “Deep-LK for Efficient Adaptive Object Tracking”, 2017
- Dymczyk et al., “Will It Last? Learning Stable Feature for Long-Term Visual Localization”, 2016
ASL has been working on fixed-wing UAVs since 2007, more recently shifting the research focus from solar powered UAVs to autonomous planes that can also operate beyond visual line-of-sight to the (safety) pilot. For example in the case of a GPS outage or failure, it is important that the UAV is able to continue its mission. Currently, ASL is able to map an area, generate a globally consistent map and localize within the map if the environmental and illumination changes are neglectable. Reliably aligning the live image stream from the drone to satellite imagery, while relatively easy for humans, poses still a challenging task for computer vision algorithms. In this context, the project will build up on recent advances in the field of deep learning, in particular in learning landmarks that are stable over time. Furthermore, ASL has access to high precision satellite data from Switzerland (covering up to 10 years) which could be used as training data.
Related Literature
- Goforth et al., “Aligning Across Large Gaps in Time”, 2018 - Wan et al., “Deep-LK for Efficient Adaptive Object Tracking”, 2017 - Dymczyk et al., “Will It Last? Learning Stable Feature for Long-Term Visual Localization”, 2016
- Literature review on vision-based localization and novel deep learning methods (e.g. learning long-term stable landmarks)
- Ground truth data collection (e.g. satellite data)
- Implementation of a deep learning enhanced localization system; e.g. as outlined in [1]
- Validation with different difficulty levels (time gap, illumination changes, seasonal changes)
- Real-time deployment on fixed-wing UAV
- Literature review on vision-based localization and novel deep learning methods (e.g. learning long-term stable landmarks) - Ground truth data collection (e.g. satellite data) - Implementation of a deep learning enhanced localization system; e.g. as outlined in [1] - Validation with different difficulty levels (time gap, illumination changes, seasonal changes) - Real-time deployment on fixed-wing UAV
- Courses and experience in computer vision and (deep) learning
- C++, python
- Courses and experience in computer vision and (deep) learning - C++, python
For more information, visit: https://docs.google.com/presentation/d/1DVAy-Jl4dDyL4uEeAYm6aD9b2YS_-oTauESIQ7nvhxE/edit?usp=sharing
Please send your CV and transcript of records to hitimo@ethz.ch
For more information, visit: https://docs.google.com/presentation/d/1DVAy-Jl4dDyL4uEeAYm6aD9b2YS_-oTauESIQ7nvhxE/edit?usp=sharing
Please send your CV and transcript of records to hitimo@ethz.ch