The mobile phone offers a unique opportunity to predict a person's behaviour automatically for advanced ubiquitous services. In this note, we analyse cellular data collected as part of the Reality Mining project and use information-theoretic concepts to answer three questions (i) What time points in the day help predict a mobile phone user's activity at another time point? (ii) What time points in history are most useful to predict his future activities? and (iii) How difficult is it to predict his activity at a given time from another user's activity at another time?

In this paper we draw a methodological connection between human routine classification and the sequence alignment problem in bioinformatics. We first observe that human days exhibit important time shifts and therefore align them for comparison prior to classification. Our technique is evaluated on bimodal data including GSM and Bluetooth information collected on mobile phones. The introduction of new alignment features is found to significantly improve the accuracy of routine classification.

The aim of human activity recognition is to identify what a user or a group of users are doing at a given point in time, for example travelling or working. Activity recognition plays an important role in mobile and ubiquitous computing both as a goal in itself and as an intermediate task in the design of advanced applications. Virtually all existing activity recognition systems for mobile phones base their predictions on location cues. This approach forces the user to disclose personal information such as her home or work area. In this paper, we present a novel activity recognition system called TRAcME (temporal recognition of activities for mobile environments) which recognises generic human activities from large windows of context, Allenpsilas temporal relations and anonymous landmarks. Unlike existing systems, TRAcME handles simultaneous activities and outputs activities which are consistent with each other at the scale of a userpsilas day.

Today's high-end mobile phones commonly include one or two digital cameras. These devices, also known as cameraphones, allow their owners to take photographs anywhere, at any time with practically no cost. As a result, many urban dwellers are photographed everyday without even being aware of it. Although in many countries, legislation recognises the right of people to veto the dissemination of their image, snapshots including recognisable passers-by often end up on photo-sharing websites. In this paper, we present a cooperative system for cameraphones which automatically anonymises faces of people photographed involuntarily. Our system, called BlurMe, uses Bluetooth awareness to inform a photographer's cameraphone when people around this photographer do not wish for their picture being taken. It then identifies subjects on the photograph and anonymises other people's faces. BlurMe was tested on 876 face regions detected on 150 real-life photographs collected from a photo-sharing website and manually labelled for subject faces. The system achieved very promising results on photographs with up to three subjects.