A Design of an Optimized ORB Accelerator for Real-Time Feature Detection

In this paper, we propose an ORB hardware accelerator design. Feature-based SIFT and SURF hardware accelerators that detect features in real-time require high hardware costs. To overcome this disadvantage, an ORB algorithm consisting of simple calculations was designed as the hardware accelerator. A comparison between the proposed ORB hardware accelerator with SIFT and SURF accelerators revealed that the block RAM usage of the former decreased by approximately 25% and 83%, and its slice logic usage decreased by approximately 40% and 68%, respectively. As approximately 9 ms were required for processing, it was verified that the proposed hardware accelerator could perform real-time feature detection.