Keyframe and inlier selection for visual SLAM

Abstract

Using stereo cameras to perform Simultaneous Localization and Mapping (SLAM) is an active area of mobile robotics research with many applications. Regardless of which SLAM algorithm is used for an application, the quality of the results depends heavily on the quality and consistency of the data going into the algorithm. In this study, a novel algorithm for inlier and keyframe selection is used to produce sets of observations that can be used to perform SLAM. Several simulations are performed using data sets captured in large outdoor environments, and the results are evaluated in terms of physical consistency, covisibility between frames, and SLAM results. The results obtained from these simulations suggest that the algorithm can be useful in the implementation of SLAM.