FPGA Acceleration of Multilevel ORB Feature Extraction for Computer Vision

This work accelerates robotic vision processing using specialised low cost, low power and high performance digital design. The design approach exploits fine grain parallelism available to hardware designs on Field Programmable Gate Arrays. State of the art image features are extracted at high frame rates, even for high definition video, allowing real time robot localisation and mapping from single or multiple cameras on mobile devices and robots. The PhD validates the approach and shows it is much faster than software implementations on conventional or graphics processors with a real time demonstration.