Visual Navigation Using a Webcam Based on Semantic Segmentation for Indoor Robots

Abstract

The realization of a moving robot that can autonomously work in an actual environment has become important. A three-dimensional dense map that was created using three-dimensional (3D) depth sensors, such as light detection and ranging (LiDAR), is popular in the research field of autonomous moving robots. However, this approach has a few disadvantages: the price of 3D sensing devices and the robustness of localization in practical scenarios with many movable obstacles. To solve this problem, this paper proposes a vision-based navigation scheme that enables autonomous movement in indoor scenes; only a webcam is used as an external sensor. The experimental results from an experiment conducted in a university building demonstrated that a robot can move around on a floor.