![]() | Only 14 pages are availabe for public view |
Abstract Real time processing of image and video applications, especially Augmented Reality (AR) systems is a demand in many computer vision applications, e.g. video surveillance, traffic management and medical imaging. Augmented Reality (AR) is divided into vision processing engine and graphics processing engine. Those two processing engines use highly computational intensive tasks such as image segmentation, object recognition, and feature tracking. The processing of those functions requires high computational power, especially when they are applied on high-definition (HD) input video frames. The processing of video and image algorithms of real time AR is not capable of running on devices efficiently with low computing power at a reasonable frame rate. The collaboration of the CPU and hardware accelerators provides an efficient solution for this problem.In this thesis, the presented approach targets offloading highlycomputational pixel processing modules from processing system (PS) toprogrammable logic (PL) taking the advantage of High Level Synthesis (HLS) tool flow to accelerate the implementation of compute-intensivepixel processing tasks of AR. Hardware accelerators for AR processingtasks are proposed. The accelerators are applied on full HD video inputstream at 1080p60 of HDMI input. The performance improvementthrough hardware acceleration decreases the CPU utilization andincreases the frame rate. |