The International Association for Automation and Robotics in Construction
Publications / 2025 Proceedings of the 42nd ISARC, Montreal, Canada
One-Shot Indoor Positioning Using 360-Degree Photos
Vision-based indoor positioning approaches are increasingly gaining attention due to their cost-effectiveness and scalability. The core principle of vision-based indoor positioning is to identify the most visually similar space to the reference space images. A major challenge in previous vision-based indoor positioning methods arises from the need for large annotated datasets to ensure robust model performance, necessitating coverage of space images from various angles and points of view. To address this limitation, we propose a method that applies feature matching to 360-degree images, enabling indoor positioning with just a single reference image per space. This eliminates the need for multiple reference images by preprocessing the input image through feature-descriptor-based alignment and cube-map projection, allowing the image to be adjusted to better match the position and shape of the reference image. Experiments conducted on six different floor plans achieved an accuracy of 72.57% using only one reference image, confirming the feasibility and efficiency of this lightweight approach to indoor positioning.
