Autonomous Vehicles, Volume 2

4Visual Place Recognition for Simultaneous Localization and Mapping

Konstantinos A. Tsintotas*, Loukas Bampis and Antonios Gasteratos

Democritus University of Thrace, School of Engineering, Vas. Sofias 12, Xanthi, Greece

Abstract

As a mobile robot, e.g., an aerial, underwater, or ground-moving vehicle, navigates through an unknown environment, it has to construct a map of its surroundings and simultaneously estimate its pose within this map. This technique is widely known in the robotics community as simultaneous localization and mapping (SLAM). During SLAM, a fundamental feature is loops’ detection, i.e., areas earlier visited by the robot, allowing consistent map generation. Due to this reason, a place recognizer is adopted, which aims to associate the current robot’s environment observation with one belonging in the map. In SLAM, visual place recognition formulates a solution, permitting loops’ detection using only the scene’s appearance. The main components of such a framework’s structure are the image processing module, the map, and the belief generator. In this chapter, the reader is initially familiarized with each part while several visual place recognition frameworks paradigms follow. The evaluation steps for measuring the system’s performance, including the most popular metrics and datasets, are also presented. Finally, their experimental results are discussed.

Keywords: Mobile robot, aerial, underwater, moving vehicle, SLAM, sensory data, visual recognition, image ...