Subsurface Infrastructure Localization for GIS Data Alignment using Semantic Segmentation

The use of augmented reality in the outdoor sector is finding more and more practical applications with better sensor technology on mobile devices. In this work, we optimize our previous setup for tracking in outdoor augmented reality (AR) applications within the field of subsurface utility engineering (SUE).
We address limitations in north estimation by compass and the semi-automatic initialization of tracking fusion that are perceived as impractical for users. An algorithm for alternative GPS based north estimation is presented. Additionally, the setup is enhanced with a continuous drift correction to automate tracking completely.
Despite the highly accurate tracking, the quality of data being worked with remains a challenge. GIS data documentation is often of poor quality due to a lack of regulation. To address this issue, we propose a solution for segmenting pipes in an open excavation site and aligning available displaced GIS models. Our approach is based on segmentation in image space and reprojection of these results into 3D space, allowing for the estimation of center points along the pipe.