Hardware-Efficient Homogenized Key-Point Selection for Visual SLAM

Homogenized key-point selection is key to achieving robust visual Simultaneous Localization and Mapping (SLAM) in autonomous agents. We present the first FPGA-based visual SLAM accelerator capable of selecting uniformly distributed key-points from video frames in real-time. Using a novel gridbased streaming hardware architecture and an efficient neighbor sorting algorithm for dynamic input sizes, our design enhances robustness and resource efficiency compared to existing methods, while meeting real-time requirements. Evaluations using challenging datasets demonstrate the advantages of our approach over state-of-the-art systems.