Imagine stepping into a smoke-filled building during an emergency, where every second counts and traditional navigation fails—yet augmented reality glasses reveal your precise location on a digital floor plan, guiding rescuers to safety. In his 2020 MSc thesis, Laurens Oostwegel pioneered a solution using the Microsoft HoloLens to combat SLAM’s drift errors through continuous spatial matching techniques like Iterative Closest Points, Instantaneous Kinematics, and Hough Transform, achieving sub-meter accuracy in most tests without any pre-installed infrastructure. As indoor positioning evolves with AI and 5G integrations in 2026, this foundational work remains a beacon, proving AR’s potential to revolutionize emergency response while highlighting the need for even more robust systems to handle real-world artifacts and uncertainties.
AR towards the future
Indoor navigation lacks the reliability of outdoor GPS, creating challenges in large buildings or during crises. This impacts emergency responders needing accurate location data to operate effectively, as well as industries where efficient movement indoors can enhance safety, productivity, and operational efficiency.
Before this study, indoor positioning typically depended on pre-installed infrastructure, including Bluetooth beacons, ultra-wideband (UWB) systems, Wi-Fi fingerprinting, or RFID tags. These approaches offered varying accuracies—UWB could achieve 0.1 meters, while Wi-Fi ranged up to 30 meters—but required advance setup, limiting their use in unplanned situations like emergency response. Augmented Reality (AR) devices, such as the Microsoft HoloLens, employed Simultaneous Localization and Mapping (SLAM) to generate positions without installations. However, SLAM positions were relative to the scanned environment, not aligned with standard floor plans or maps, reducing contextual usefulness. Additionally, SLAM algorithms accumulated drift errors over time, compromising long-term accuracy in dynamic indoor settings.
The Research
The thesis investigates: How can the Microsoft HoloLens improve indoor positioning by integrating its real-time mesh with an existing floor plan? The goal is to develop a continuous alignment method using spatial matching to correct SLAM drift and reference positions to floor plans. This differs from traditional methods by eliminating the need for pre-installed hardware, emphasizing 2D registration for practical application in emergency response, and evaluating three techniques—Iterative Closest Points (ICP), Instantaneous Kinematics (IK), and Hough Transform (HT)—for performance.
The research proposes a method to address SLAM’s challenges by continuously registering the HoloLens’ 3D mesh to a 2D floor plan, converting relative positions into those referenced against a map. It involves extracting vertical walls from the mesh for 2D projection and applying spatial matching algorithms: ICP, which iteratively minimizes point-to-line distances using least squares; IK, an ICP variant incorporating velocity vectors for handling small adjustments; and HT, which uses Hough space properties where rotation is independent of translation and scale. An initial manual alignment is provided, with automated updates every 15 seconds to mitigate drift. Preprocessing steps include clipping floor plans and filtering mesh data. This approach allows positioning without infrastructure, tested in real buildings, and offers insights into algorithm trade-offs for engineers working on AR-based localization.
Research Results
Testing in multiple buildings revealed that the Hough Transform provided the highest accuracy and fastest computation, maintaining average errors below 1 meter in 80% of experiments, with peaks up to 5 meters. In 20% of cases, errors exceeded 100 meters, attributed to scan quality and artifacts like furniture. Uncorrected HoloLens drift reached 18 meters over 350-meter paths, but successful registrations reduced this notably. IK and ICP showed lower performance, with IK sensitive to certain edge cases.
This method supports infrastructure-free indoor positioning, potentially aiding emergency responders and sectors requiring flexible navigation. Applications could extend to multi-story buildings or dynamic environments. Further work might focus on improved outlier detection, integration with 3D models, or adaptations for challenging conditions like smoke, building on the thesis’s findings.
Looking Forward
As indoor environments grow more complex, reliable positioning without setup remains a key engineering challenge—this work highlights AR’s role in addressing it, though robustness improvements are needed. As noted in the thesis, “the results are promising and do open the door for indoor navigation using Augmented Reality.” We appreciate Laurens Oostwegel’s contribution; for collaboration or insights, contact via Delft University. Reference: Oostwegel, L. (2020). Indoor Positioning using Augmented Reality. MSc Thesis, Delft University of Technology.