A new real-time visual SLAM algorithm based on the improved FAST features

Abstract

The visual SLAM is less dependent on hardware, so it attracts growing interests. However, the visual SLAM, especially the Extend Kalman Filter-based monocular SLAM is computational expensive, and is hard to fulfill real-time process. In this paper, we propose an algorithm, which uses the binary robust independent elementary Features descriptor to describe the features from accelerated segment test feature aiming at improving feature points extraction and matching, and combines with the 1-point random sample consensus strategy to speedup the EKF-based visual SLAM. The proposed algorithm can improve the robustness of the EKF-based visual SLAM and make it operate in real-time. Experimental results validate the proposed algorithm. © 2013 Springer-Verlag.