In this thesis project you will be working with Veoneer in Linköping, in collaboration with the division for Media and Information Technology at Campus Norrköping.
Modern camera-based vehicle safety systems rely heavily on machine learning and consequently require large amounts of training data to perform reliably. One significant problem with this is that data containing the most important scenarios, e.g. potential accidents, is often the most difficult to collect. One way to alleviate the problem is to instead make use of synthetic data for these scenarios.
The goal of this master's thesis will be to investigate to what extent synthetic data can replace real data for specific critical scenarios. Since only a limited amount of data consisting of such real world scenarios is available, the recommended approach is instead to study proxy scenarios.
One possible example would be to study a pedestrian detector trained on data where all samples of pedestrians walking closely in front of the vehicle (e.g. at a crosswalk) have been removed. This detector could then be compared to other detectors trained on datasets where the removed samples have been replaced by various degrees of synthetic data. All detectors are evaluated on the available real data where the proxy scenarios haven't been excluded.
To generate synthetic scenarios, it's suggested to use CARLA, an autonomous driving simulation platform.
Investigate to what extent synthetic data can replace real data for specific critical scenarios.
Students with a background in computer science, electrical engineering applied mathematics or similar, with experience from courses in computer vision, machine learning and in particular deep learning approaches to computer vision tasks
Contact: Gabriel Eilertsen (email@example.com), division of Media and Information Technology, Department of Science and Technology, Campus Norrköping, Linköping University