Fine-grained activity recognition of pedestrians travelling by subway

Abstract

With the now widespread usage of increasingly powerful smartphones, pro-active, context-aware, and thereby unobstrusive applications have become possible. A user’s current activity is a primary piece of contextual information, and especially in urban areas, a user’s current mode of transport is an important part of her activity. A lot of research has been conducted on automatically recognizing different means of transport, but up to know, no attempt has been made to perform a fine-grained classification of different activities related to travelling by local public transport.In this work, we present an approach to recognize 17 different activities related to travelling by subway. We use only the sensor technology available in modern mobile phones and achieve a high classification accuracy of over 90%, without requiring a specific carrying position of the device. We discuss the usefulness of different sensors and computed features, and identify individual characteristics of the considered activities.