Quadrant segmentation and ring-like searching based FPGA implementation of ORB matching system for Full-HD video

Full-HD video has drawn more and more attention in advanced computer vision applications which rely on more details in image. Benefit from high resolution input, local feature based matching system which at base of various vision applications, can also get better performance due to more available information. However, high resolution brings massive data and makes it challenging to achieve real-time and low cost at the same time. This paper proposes an ORB-based matching system for Full-HD video which implemented on FPGA. To improve nonlinear functions and feature steering part of ORB in hardware, the Quadrant Segmentation based orientation detector and Ring-like Searching based feature steering are proposed to make original operation more suitable for hardware. Evaluation shows that the proposed ORB matching system can complete feature extraction and matching for one Full-HD(1920×1080) image within 13.37ms and save almost 75% resources on average in feature extraction part compared with SIFT-based design.