Camera oscillation pattern for VSLAM: Translational versus rotational

Abstract

Visual SLAM algorithms exploit natural scene features to infer the camera motion and build a map of the environment landmarks. SLAM algorithm has two interrelated processes localization and mapping. For accurate localization, we need the features location estimates to converge quickly. On the other hand, to build an accurate map, we need accurate localization. Recently, a biologically inspired approach exploits deliberate camera oscillation has been used to improve the convergence speed of depth estimate. In this paper, we explore the effect of camera oscillation pattern on the accuracy of VSLAM. Two main oscillation patterns are used for distance estimation: translational and rotational. Experiments, using static and moving robot, are made to explore the effect of these oscillation patterns on the VSLAM performance.