Robust vSLAM for Dynamic Scenes

This paper proposes a robust visual Simultaneous Localization and Mapping(vSLAM) method that consists of two parts: (1)an initialization method for a 3D map and camera poses, and (2)an outlier rejection method for moving objects. We introduce weighted tentative initial values of a camera pose and 3-D map to reduce the user operation load. We also distinguish outliers of feature points between moving objects and errors to stably estimate camera poses in dynamic scenes. To achieve this, we first construct an angle histogram based on outlier flows at a current frame, then approximate the obtained angle histogram using a mixture of Gaussian functions. Finally, we estimate the parameters for Gaussian mixtures using the ExpectationMaximization algorithm. Results are shown to illustrate the superior performance of the proposed method.