Overview

The ubiquity of mobile communication devices such as smartphones enables the emergence of context-aware applications and services that pro-actively respond to specific user activities or situations. Context information, i.e., the specific state each user is in, allows communication providers to develop and thus offer new, added-value, services for a wide range of applications such as social networking, advertising, navigation or leisure. Of growing importance are health-related services and applications that rely on the accurate detection of each user’s physical activity either at specific instances or throughout days or even weeks. Using this information it is possible to discover and analyze physical activity patterns and, e.g., help individuals to lead healthier life-styles.

At the core of these applications/services is the ability of mobile devices to accurately detect specific physical settings or user contexts, using either internal or external sensors such as accelerometers, GPS, light, image or sound.  Accurately capturing and managing the wealth of sensor data and managing the resulting extracted context information in a meaningful way in the perspective of applications are a substantial technical challenge. If not adequately preprocessed, the raw data these sensors generate, in terms of raw bandwidth, storage and energy consumption, can overwhelm even contemporary high-end mobile devices. Equally critical is the need of low energy consumption and high accuracy in context detection and management of communication, as the required continuous operation can quickly drain mobile energy supplies. False classification of contexts will eventually lead to inconsistent activity monitoring patterns making the development of value-added applications increasingly complex and frustrating to undertake.

It will contribute to advances within the priority axis with new algorithms and middleware techniques to support the development of mobile context-aware applications. The new algorithms and middleware will be integrated in prototype systems to be used by all parties involved in the project. The context and social network information provided by the techniques to be researched and developed will be used for automatic user profiling and can be used for supporting value-added services that depend on user context (e.g., content recommendation and targeted marketing are two direct applications that can benefit from user profiling). The proposed project addresses the key challenges and enabling techniques for context-aware application development in three major aspects. Firstly, we propose to research and develop activity detection techniques focusing on low energy, and high accuracy. Secondly, we will explore context aggregation approaches and algorithms based on statistical classification, to uncover activity patterns. Lastly, as mobile systems interface will undoubtedly evolve, we will develop a middleware and domain-specific programming environment for the rapid prototyping of context-aware applications. As part of the proposed project we will evaluate the techniques using a real prototype consisting of a smartphone and an array of sensors mounted on a personal vest inspired in the prototype system previously developed by members of our team.

The research efforts on classification techniques will be focused on context-inference to be implemented in mobile devices with limited capabilities. The existence of instances without target labels will lead, necessarily, the research to the area of semi-supervised classification. Additionally, the techniques need to deal with data streams and with small windows of sensing data. Furthermore, these techniques should be energy-aware and should manage multiple configurations to save energy, an important aspect in mobile devices. The objective is to minimize the classification error. Specifically, we will address energy-aware algorithms for the detection of physical activities based on accelerometer and other sensor data. The outcome of these algorithms will be integrated with a high-level algorithm for the detection of high-level user contexts and related activities. To support the embedded software development of context-aware applications, we will continue the development of a flexible middleware and programming level environment. This environment will facilitate the integration of new sensors, energy saving strategies, and the definition of added-value services. We will use a smartphone based prototype system extended to a manysensory module to evaluate these algorithms in real life scenarios. Our ultimate goal is to research and analyze techniques for the advanced wearable sensory technologies that might permeate user’s daily-life in a near future.

The middleware and the techniques R&D during the project will be validated in an industrial context. An use case provided by Altice Labs will allow us to estimate the feasibility of the CONTEXTWA approach in a real context and to ultimately evaluate the possibility for technology transfer.