YorkU Indoor Localization Dataset

A longitudinal multi-building indoor localization dataset with AR-verified centimeter-accurate ground truth. Designed for evaluating WiFi fingerprinting, sensor fusion, and temporal robustness of indoor positioning systems.

Get Dataset Cite Work
4
Buildings
28,822
Sq Meters
9,659
Fingerprints
1.8 cm
GT Accuracy

Data Collection Methodology

Centimeter-level accuracy achieved via Augmented Reality and Visual-Inertial SLAM.

AR-Based Ground Truth System

  • Hardware: Microsoft HoloLens 2 with depth sensing (1 MP TOF, 120°×120° FOV) and visual-inertial SLAM.
  • Persistence: Spatial anchors persisted via Azure Spatial Anchors (cloud-based, cross-session).
  • Alignment: MRKit toolkit used for floor plan alignment with CAD drawings.
  • Protocol: Two-operator systematic approach. HoloLens operator for spatial guidance, Android operator (Galaxy S21) running GetSensorData app.
  • Sampling: 30-second stationary collection yielding 10-15 WiFi scans per reference point.
  • Validation: Anchor accuracy validated at mean 1.8 cm (95th percentile: 3.7 cm, max: 6.2 cm) against a Leica TS15 total station at 80 points.

Four-Building Testbed

Diverse architectural styles, construction materials, and environmental challenges.

PSE

Petrie Science & Engineering

12,544 m² 89 APs 312 Ref Points
  • • 56×56 m symmetric square floor plan
  • • Reinforced concrete construction
  • • Square corridors, exterior vs interior labs
Key Challenge: Perceptual aliasing from symmetry, magnetic anomalies near corner staircases (>20 µT).

BCE

Bergeron Centre for Engineering

13,800 m² 67 APs 289 Ref Points
  • • 46×60 m rectangular footprint, LEED Gold
  • • Steel frame with glass curtain walls
  • • Labs connected by irregular corridors/open areas
Key Challenge: Severe RF multipath (path loss exponents 2.8–4.2) and signal dead zones.

STC

Student Teacher Centre

702 m² 45 APs 87 Ref Points
  • • 78×9 m linear floor plan, open on one side
  • • Mixed commercial construction
  • • Primary pedestrian thoroughfare
Key Challenge: Extreme occupancy variation (10x between quiet/peak), lighting changes (100–2000 lux).

LRC

Lassonde Research Centre

1,776 m² 38 APs 159 Ref Points
  • • 74×24 m rectangular footprint
  • • Specialized research facility construction
  • • Closed-off building, windows only at entrance
Key Challenge: Dense research equipment creating strong magnetic signatures (>50 µT perturbations).

109-Dimensional Sensor Modalities

Integrated multi-sensor fingerprints collected simultaneously.

WiFi RSSI

100 Dimensions
  • • IEEE 802.11 (2.4 & 5 GHz)
  • • 0.5 Hz scan rate
  • • 8–40+ visible APs per location
  • • Range: -85 to -35 dBm
  • • EMA filtered (α=0.7)
  • • Zero-padded normalization

Inertial (IMU)

6 Dimensions
  • • 3-axis accelerometer
  • • 3-axis gyroscope
  • • 100 Hz sampling rate
  • • Motion state detection
  • • Capable of step counting

Magnetometer

3 Dimensions
  • • 3-axis magnetic field sensor
  • • 50 Hz sampling rate
  • • Hard & soft iron calibrated
  • • Toronto baseline: ~48 µT
  • • Captures structural signatures

Longitudinal Evolution
2021 → 2022

A core value of this dataset is capturing how indoor environments degrade localization performance naturally over a full calendar year.

Infrastructure Churn

~59 new APs installed and 12 decommissioned — a net gain of 47 (24.5%, 192 → 239). Only ~78% of access points remained stable.

Physical Alterations

Partial drywall renovations in BCE, new electrical panels, and ~23% of office areas underwent furniture rearrangements.

Magnetic Fingerprint Stability

While WiFi correlation dropped (0.70–0.85), magnetic fingerprints remained highly stable (>0.95 correlation YoY) except where new MRI machines were installed in LRC.

Data Distribution by Year

Fall 2021 Campaign 4,480 samples
Fall 2022 Campaign 5,179 samples
HoloLens Anchor Drift (Annual): < 0.5% (mean 1.8 cm)

Dataset Format

Data is provided in CSV format, organized by building (PSE, BCE, STC, LRC) and by year (2021, 2022).

  • Feature Vector: 109 dimensions (100 WiFi + 6 IMU + 3 Magnetometer)
  • Target: 2D coordinates (X, Y) in building-local frame
  • Metadata: Anchor IDs, timestamps, session IDs, building/floor identifiers

Licensed under CC BY 4.0.

Citation

If you use this dataset in your research, please cite it as follows: 

@misc{ahmad2026indoor,
  author       = {Ahmad, Afnan and Sohn, Gunho},
  title        = {YorkU Indoor Localization Dataset},
  year         = {2026},
  howpublished = {\url{https://yorkuindoor.ausmlab.com}},
  note         = {AUSMLab, York University},
}