GPS/Pedometer

Problem

• When GPS goes blind, navigation doesn’t just degrade, it collapses. From multi-storey buildings and underground transport to disaster zones and dense urban canyons, modern navigation systems fail precisely where reliability matters most

• The smartphone in your pocket (or that you're using to read this) already carries the sensors needed to solve this. The challenge is turning noisy, drifting inertial data into a coherent, trustworthy notion of “where am I?”

Why it matters

• GPS-denied environments expose a fundamental gap between where navigation systems fail and where reliable position information is most needed. This motivates sensor-based solutions that can deliver continuous, self-contained positioning without external infrastructure 

Approach

• Exploited commodity smartphone sensors (accelerometer, gyroscope, magnetometer) to construct a fully self-contained indoor positioning system operating without external references 

• Decomposed positioning into heading and distance estimation, enabling modular development, calibration, and validation of each subsystem 

• Applied signal processing and dead-reckoning techniques to extract motion primitives (steps, angular displacement) from noisy inertial measurements 

• Fused complementary sensors using a Kalman Filter, combining short-term gyroscope accuracy with long-term magnetometer stability 

Key Insight

• Framed heading estimation as a stochastic state estimation problem, enabling principled sensor fusion rather than heuristic averaging 

• Quantified sensor trust using measured noise statistics, allowing the estimator to dynamically weight gyroscope and magnetometer data based on uncertainty

• Leveraged a Discrete Fourier Transform to separate walking dynamics from drift and noise, improving robustness of step detection and distance estimation

Result

• Kalman-filtered heading estimates consistently outperform single-sensor approaches, producing smoother and more reliable orientation estimates 

• End-to-end positioning demonstrates feasibility but exposes accumulated error, underscoring the sensitivity of dead-reckoning systems to modelling assumptions and calibration quality