Indoor navigation systems have become highly important in the case of large buildings and complexes; yet, existing solutions are inclined to be based on specialized hardware infrastructure like WiFi and Bluetooth Low Energy (BLE). This paper presents a new approach to indoor navigation without the requirement of additional hardware through the use of computer vision and augmented reality technology. The system presented makes use of COLMAP - Structure from Motion (SfM) for visualizing the indoor area from video footage. Users simply upload an initial view image to estimate their initial pose through COLMAP-based estimation of camera parameters, achieving localization errors of less than \(0.3\,\textrm{m}\) in controlled experiments. Waypoints between the starting and destination coordinates, calculated over the dense point cloud using A* search on a 3D graph, align with ground-truth paths with an average spatial error of less than \(0.25\,\textrm{m}\) . Transformation of these waypoints to ARCore space reveals spatial misalignment and jitter, which makes in-situ testing unreliable. Our vision-based pipeline, unlike BLE-based systems (which typically exhibit 1–2 m errors), performs well in the initial stages but requires improvements to be reliable for AR visualization. Key constraints include AR marker stability under dynamic lighting and occlusion, which we identify as important areas for future work.