This project demonstrates an approach to detect common smart phone positions, including “In Hand”, “Side Pocket”, “In Handbag”, and “On Table (Idle)”. We apply sensor data fusion on a set of hardware- and software-based Android sensors. As a result, data entries containing sensor fusion features are classified individually, without looking at patterns. The overall system design and accuracy are presented using a Google Nexus 6 device.
The neural network used for classification. Nodes: 13 - 8 - 8 - 4
Prediction system diagram. Remote prediction using Weka.
Android Application. Left: Logging mode, Right: prediction mode.