The goal of this thesis is to implement a Feature Detector (ex. Harris, Fast, SUSAN, LoG, DoG), Descriptor (SURF, BRISK, BRIEF, ORB, FREAK) and Matcher/Tracker using the nVidia Cuda Framework for high-efficiency real-time execution on a nVidia Jetson TX2 computer. Focus should be laid on exploiting new computational architectures arising from Machine Learning, such as fast convolutions and GPU-aided computation.
The goal of this thesis is to implement a Feature Detector (ex. Harris, Fast, SUSAN, LoG, DoG), Descriptor (SURF, BRISK, BRIEF, ORB, FREAK) and Matcher/Tracker using the nVidia Cuda Framework for high-efficiency real-time execution on a nVidia Jetson TX2 computer. Focus should be laid on exploiting new computational architectures arising from Machine Learning, such as fast convolutions and GPU-aided computation.
The thesis should provide a comparison of execution speed with respect to execution on an ARM CPU. The end goal would be a clean implementation for future use in VIO pipelines with the possibility of a theoretical contribution if any improvements to the state-of-the-art algorithms is found.
The thesis should provide a comparison of execution speed with respect to execution on an ARM CPU. The end goal would be a clean implementation for future use in VIO pipelines with the possibility of a theoretical contribution if any improvements to the state-of-the-art algorithms is found.