Recognition of high-level activities with a smartphone

The recognition of high-level activities (such as work, transport and exercise) with a smartphone is a poorly explored topic. This paper presents an approach to such activity recognition that relies on the user's location, physical activity, ambient sound and other features extracted from smartphone sensors. It works in a user-independent fashion, but can also take advantage of activities labeled by the user. It was evaluated on a real-life dataset consisting of ten weeks of recordings. While most activities were recognized quite accurately, the recognition of some revealed two challenges of recognizing diverse lifestyle activities: the ambiguity of some activities, and the inadequacy of smartphone sensors for others.