Drift-free BIM Alignment for Mixed Reality Visualization through Image Style Transfer and Feature Matching

Abstract This research introduces a novel methodology that automates the precise alignment between real and virtual environments in Mixed Reality (MR) applications, specifically tailored for the construction industry. A significant challenge in MR systems is the accumulation of camera pose estimation errors, leading to trajectory drift and reduced localization accuracy over time. Our approach addresses this by integrating HoloLens' spatial mapping capabilities with Image style transfer and geometric feature matching, enabling robust alignment between real-world HoloLens images and Building Information Modeling (BIM). By bridging the visual domain gap through image style transfer, we enhance feature correspondence, effectively eliminating drift errors that accumulate during device movement. A comprehensive evaluation using 1,408 image pairs demonstrates improved localization accuracy and reliable alignment of BIM in the real world for enhancing efficiency in the construction industry.