Visual-SLAM Based 3-Dimensional Modelling of Indoor Environments

Abstract

Simultaneous localization and mapping (SLAM) is used in many fields to enable robots to map their surroundings and locate themselves in new circumstances. Visual-SLAM (VSLAM), which uses a camera sensor, and LiDAR-SLAM, which uses a light detection and ranging (LiDAR) sensor, are the most prevalent SLAM methods. Thanks to its benefits, including low-cost compared to LiDAR, low energy consumption, durability, and extensive environmental data, VSLAM is currently attracting much attention. This study aims to produce a three-dimensional (3D) model of an indoor environment using image data captured by the stereo camera located on the Unmanned Ground Vehicle (UGV). Easily measured objects from the field of operation were chosen to assess the generated model's accuracy. The actual dimensions of the objects were measured, and these values were compared to those derived from the VSLAM-based 3D model. When the data were evaluated, it was found that the size of the object produced from the model could be varied by +/- 2cm. The surface accuracy of the 3D model produced has also been analysed. For this investigation, areas where the walls and floor surfaces were flat in the field were selected, and the plane accuracy of these areas was analysed. The plain accuracy values of the specified surfaces were determined to be below +/- 1cm.