Requirements before choosing an approach
Sensor fusion is the practice of combining readings from two or more sensor types to produce a more reliable position estimate than any single source could deliver on its own. In indoor navigation, this matters because no individual technology behaves consistently across all environments, devices and visitor behaviours.
A smartphone walking through a venue can draw on Bluetooth beacons, Wi-Fi access points, its own inertial measurement unit (IMU), magnetometer, barometer and, in some cases, camera or lidar. Each of these has distinct strengths and failure modes. Beacon RSSI fluctuates with interference and human bodies. Wi-Fi fingerprinting degrades when access points are moved or reconfigured. IMU drift accumulates over time without external correction. Sensor fusion algorithms — commonly Kalman filters, particle filters or simpler weighted averaging — take these imperfect inputs and reconcile them into a single estimated position.
The logic is straightforward: when one signal becomes unreliable, others compensate. A visitor passes a metal-framed doorway that disrupts the magnetometer; beacon and Wi-Fi data carry the position through that moment. A beacon battery dies; IMU dead-reckoning bridges the gap until the next beacon is reached. The fusion layer decides how much to trust each input at any given moment, typically based on signal quality metrics, known environmental characteristics and historical consistency.
How Fusion Differs from Simple Triangulation
Triangulation or trilateration using beacons alone assumes you can convert signal strength into distance reliably. In practice, walls, fixtures and crowd density make that conversion unstable. Sensor fusion does not try to perfect that conversion; instead, it treats the beacon reading as one probabilistic input among several, and lets the algorithm downgrade its influence when the signal looks anomalous. This is a fundamentally different approach from trying to improve a single technology's raw accuracy.
Where Fusion Happens
Fusion can run on the device, on a server, or both. On-device fusion is more common in visitor-facing navigation because it reduces latency and keeps raw sensor data on the phone, which has privacy advantages. Server-side fusion can incorporate inputs the device cannot see — such as fixed cameras or ceiling-mounted UWB anchors — but requires a live data connection and raises more immediate privacy questions. Hybrid approaches push lightweight fusion to the device and send only aggregated position updates to the server for analytics or coordination.
Coordinate technology, people and process
Museums and Galleries
Museum visitors move slowly, pause frequently and often follow predictable routes. Fusion here typically combines sparse beacon placement with IMU step counting and occasional Wi-Fi fixes. The barometer helps resolve floor level in multi-storey buildings. Because the movement pattern is gentle, IMU drift is manageable between beacon corrections, and the system can tolerate fewer reference points than a fast-moving retail environment would require. The practical benefit is that exhibit-level triggering remains reliable even when a single beacon is obscured by a large display case or a group of visitors.
Retail Environments
Shoppers move more erratically, change direction abruptly and spend variable time in each zone. Fusion in retail often leans more heavily on Wi-Fi fingerprinting alongside beacons, because Wi-Fi infrastructure is usually already present and provides a coarse position quickly. IMU data smooths the path between Wi-Fi updates. The key operational consideration is that Wi-Fi access point positions must be kept current in the system; a single relocated access point can corrupt the fingerprint map and degrade fusion quality until the map is updated.
Events and Temporary Venues
Pop-up events present a particular challenge: the environment changes rapidly, there is little time for extensive fingerprinting, and infrastructure is temporary. Fusion approaches here tend to prioritise beacon-and-IMU combinations that can be calibrated quickly, often using a walk-through survey on the day of setup. The trade-off is lower absolute accuracy in exchange for faster deployment. Event organisers should expect to re-calibrate if layout changes occur between setup day and the event opening.
Accessibility Wayfinding
For visitors who rely on turn-by-turn indoor directions — for example, wheelchair users or visually impaired visitors — fusion stability matters more than peak accuracy. A position estimate that jumps between two aisles is worse than one that is consistently a metre or two off but smooth. Fusion algorithms tuned for accessibility use cases typically apply stronger smoothing and are more conservative about accepting sudden position changes, even if that means slightly slower response to genuine direction changes.
Calibration and Environmental Surveys
Sensor fusion does not eliminate the need for site calibration. Wi-Fi fingerprinting requires a radio map. Beacon-based fusion requires measured RSSI-to-distance relationships for the specific environment. IMU integration requires step-length calibration that varies with user height and walking speed. The difference is that fusion can tolerate lower calibration precision in any single layer, because the other layers compensate. However, the cumulative effect of poor calibration across all inputs will still produce unreliable results.
Evidence for scaling or stopping
Assuming Fusion Fixes Bad Underlying Data
The most frequent error is treating sensor fusion as a remedy for inadequate infrastructure. If beacon placement is too sparse, Wi-Fi maps are outdated, or the venue has areas with no detectable signals at all, the fusion algorithm has nothing useful to combine. Fusion improves good data; it does not manufacture data where none exists. Before investing in fusion logic, verify that each underlying sensor layer meets a minimum standard of coverage and reliability in the spaces that matter.
Ignoring Device Heterogeneity
Not all smartphones have the same sensors, sensor quality or firmware behaviour. Budget devices may have noisier IMUs, less sensitive Bluetooth radios or barometers that report at lower resolution. A fusion algorithm tuned on high-end handsets during a pilot may perform noticeably worse on the devices most visitors actually carry. Practical checks include testing on a representative mix of common devices, not just the project team's phones, and building in device-type detection where the fusion platform supports it.
Over-Engineering for the Use Case
Full multi-sensor fusion with Kalman filtering, floor detection and magnetic matching is appropriate for large hospitals or airports where sub-metre accuracy and multi-floor routing are genuinely needed. A single-storey retail unit where the objective is zone-level triggering — "customer is in the footwear department" — rarely justifies that complexity. The operational cost of maintaining a sophisticated fusion system, including calibration updates and algorithm tuning, often exceeds the benefit for simple use cases. Match the fusion approach to the decision the system needs to support.
Battery and Computational Overhead
Continuously sampling the IMU, Bluetooth radio and Wi-Fi adapter drains the visitor's battery. If the navigation experience noticeably reduces phone battery life, visitors will disable Bluetooth, close the app or avoid using it altogether. Practical mitigation includes adaptive sampling — increasing sensor polling frequency only when the user is actively navigating, and reducing it when stationary — and being transparent about battery impact during onboarding.
Privacy Implications of Multi-Sensor Data
Combining multiple sensor streams produces a richer picture of a person's movement than any single source. Under UK GDPR and the ICO's guidance on location data, this raises questions about data minimisation and purpose limitation. If the fusion system processes magnetometer readings that could reveal which specific display case someone stood in front of, that is more granular than a simple zone entry log. Organisations should define the minimum position granularity needed for the stated purpose, document why finer detail is not retained, and ensure the fusion pipeline does not silently collect or store data beyond what the privacy notice describes.
Key Checks Before Committing to a Fusion Approach
- Confirm each underlying sensor layer has been surveyed and calibrated independently before relying on fusion to mask weaknesses.
- Test on at least four to six different device models that represent your actual visitor base.
- Verify that the fusion platform can adapt its weighting when a sensor type becomes unavailable — for example, when a visitor disables Wi-Fi.
- Ask the supplier what happens to accuracy when the user puts their phone in a pocket or bag, as this affects IMU and Bluetooth reception differently.
- Review the privacy notice and data-flow documentation to confirm that fused position data is not retained at higher granularity than disclosed.
- Establish a maintenance schedule for updating Wi-Fi fingerprint maps and beacon calibration as the physical space changes.
