Visual-based semantic simultaneous localization and mapping for Robotic applications: A review

Abstract

One of most important techniques that plays a key role in elevating a mobile robot’s independence is its ability to construct a map from an unknown surrounding in an unknown initial position, and with the use of onboard sensors, localize itself in this map. This technique is called simultaneous localization and mapping or SLAM. Over the last 30 years, numerous new and interesting inquiries have been raised, with the improvement of new techniques, new computational instruments, and new sensors. However, the big challenges facing mobile robots in the next decade, as in the autonomous urban vehicles, require extended representations that exceed traditional mapping found in classical SLAM systems, i.e. the so-called semantic representation. The main goal of a SLAM system with semantic concepts is to expand mobile robots’ services and strengthen human-robot interaction. Related works reviewed show that the visual-based SLAM or VSLAM has received a great deal of interest in the last decade. This is due to the visual sensors’ capability to provide information of the scene whereas they are low-priced, smaller and lighter than other sensors. Unlike the metric representation, semantic mapping is still immature, and it comes up short on durable formulation. This paper aims to systematically review recent researches related to the semantic VSLAM, including its types, approaches, and challenges. The paper also deals with the classical SLAM system by giving an overview of necessary information before getting into detail. This review also provides new researches in the SLAM domain facilities to further understand the anatomy of modern VSLAM generation, discover recent algorithms, and apprehend some open challenges.