Investigation of Visual Reconstruction Techniques Using Mobile Robots in Confined Environments

Abstract

The use of mobile robots to explore and supervise confined and uneven environments such as pipes, caves, and galleries improves operational safety by removing human operators from these dangerous areas. In many types of inspections, the robot must generate realistic, colored, and geometrically accurate maps, which experts can use to study and assess the environment remotely at a safe location. This paper investigates two approaches for visual reconstruction of confined environments: a point cloud registration combined with visual odometry and the RTAB-Map SLAM method. Real experiments performed inside a closed corridor and within an underground gold mine show that both approaches are suitable for estimating the robot's pose and perform 3D mapping. Preliminary results indicate that the point cloud registration generates denser maps suitable for visual inspection, and RTAB-Map provides less noisy maps proper for navigation.