Real-time Route Inference and Learning for Smartphone Users using Probabilistic Graphical Models

Smartphones are equipped with a rich set of embedded sensors such as accelerometer, GPS, digital compass, microphone and camera. And most smartphone users are carrying them outdoor at all times. Sensor-logged datasets contain user-specific activity patterns, which enable modeling personalized usual lives. Here, we propose a real-time route inference and learning method using dynamic Bayesian networks(DBNs) and Rao-Blackwellized particle filtering (RBPF) given sequential observed values. For experimental verification, we gathered personal sensor data of GPS, accelerometer and action recognizer for 64 days and extracted significant places and routes from them. And we predicted traveling destinations and routes probabilistically. The experimental results showed that the proposed model represents the intention of our subject properly.