Heading Estimation via AI

A machine learning approach to pedestrian heading estimation — inferring the direction a person is walking from sensor data, without requiring GPS line-of-sight or explicit user input.

Problem

Accurate pedestrian heading estimation is a core challenge in indoor navigation, pedestrian dead reckoning, and proximity-aware applications. Traditional compass-based approaches are disrupted by magnetic interference in built environments.

Approach

The project explored a neural network approach to fusing accelerometer, gyroscope, and barometric data to infer heading direction. Training data was collected through instrumented walks, and the model was evaluated against ground truth from GPS-rich outdoor environments.

This work was connected to the broader Pollination Networks research on distributed sensing and location-aware networks.