Accelerometer and gyroscope data-based detection of on-the-same-body wearables

Details

Supervisors

Lee, John

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique

Abstract

The number of wearable devices grows rapidly in our lives for many applications. They usually must be connected to collect data and transmit them to a device to compute statistical results. Unfortunately this connection is provided by manual or intrusive techniques of pairing that can lead to a mislabeling data if people walking together exchange unintentionally their wearables but are always paired to their own computational device (i.e. a smart phone).To avoid this kind of situation, I provide a generalized method based on accelerometer and gyroscope as we can expect they are embedded in each wearable. I extract features from the accelerometer and gyroscope and use the coherence to determine how well features correlate for different locations on the body. I train two classifier : a random forest and a support vector machine with these feature coherences to recognize whether two sensors are on the same body. The evaluation of the method is made on a unbalanced dataset collected on five individual people walking individually with sensors on five different locations. The generalized method achieve a balanced classification rate over 90%.