RGB-D SLAM of Autonomous Mobile Robot Toward the Dynamic Environment

Abstract

Simultaneous Localization and Mapping (SLAM) is widely used in Autonomous Mobile Robot (AMR) applications. Most SLAM methods are applied in a static environment, which significantly limits their application in real-world settings. This study proposes an RGB-D SLAM based on ORB-SLAM2 that is more robust to dynamic environments and is implemented in real-time. Object detection and multi-view geometry is used to detect and prevent moving objects impacting system, our SLAM system without the need for special hardware such as GPU. In order to verify the proposed method, this study conducts experiments using the public TUM dataset and real environment also compares the system time cost with that for other SLAM applications in dynamic environments.