FCH-SLAM: A SLAM Method for Dynamic Environments using Semantic Segmentation

Abstract

Static environments are a prerequisite for most visual simultaneous localization and mapping (SLAM) systems because the dynamic matching points from moving objects in the camera’s field of view interrupt the localization process. The noise of the dynamic objects also contaminates the constructed maps. In this study, we propose a SLAM system designed to reduce the effects on the accuracy caused by dynamic objects to solve this issue. The noise points of dynamic objects are removed by combining depth information and semantic information. We evaluated the proposed method on the TUM RGB-D dataset, and the experimental results show that it performed well in dynamic environments, obtaining a high accuracy in most situations with a relatively high processing speed.