Advancement of vision-based SLAM from static to dynamic environments

Abstract

Simultaneous Localization And Mapping (SLAM) allows a mobile robot to construct a map of an unknown, static environment and simultaneously localize itself. Real world environments, however, have dynamic objects such as people, doors that open and close, movable furniture, cars, and other robots. In order for a mobile robot to be completely autonomous in such an environment, the SLAM algorithm must be complemented by algorithms that can manage these dynamic objects. Moving objects must be not be included in the SLAM map as they may lead to localization errors and reduce map quality. Recent years, have seen the advancement of vision-based SLAM from static to dynamic environments, where SLAM coupled with Detection And Tracking of Moving Objects (DATMO) algorithms is able to handle moving objects. This paper provides a literature review of the advancement of vision-based SLAM from stationary to dynamic environments. It looks at the algorithms and sensor configurations that have enabled mobile robots to produce a reliable map of the environment for localization and task completion in the presence of moving objects. It determines the algorithms that will be used to allow a mobile robot to perform SLAM and DATMO with two RGB-D cameras in an indoor dynamic environment and concludes where a contribution can be made for SLAM and DATMO.