This comprehensive and detailed review of existing approaches and techniques for opportunistic user context recognition concentrates on methods that operate autonomously. The approach proposed in this paper focuses on mobile phone-centric methodologies.

The analysis follows three main process stages: sensing, preprocessing, and context recognition. For the sensing stage, the authors explore three main types of sensors--inertial, positioning, and ambient--and describe how they work, the types of information they can provide, their strengths and weaknesses, and how they have been applied in context recognition systems for mobile phones.

The section devoted to preprocessing presents different approaches for filtering and converting raw sensor data into a finite set of features for further analysis. After a discussion of calibration methods to address the effects of variable device position and orientation on the measurements provided by sensors, the authors describe the types of features that can be generated from raw data (time domain, frequency domain, and heuristic features), organized around the types of contexts that can be inferred from them (user physical activity, social interactions, and environment).

Finally, the classification algorithms that have been employed to infer the user context from the input features on mobile devices are analyzed. Five discriminative models and three generative models are described; their advantages, drawbacks, and challenges are highlighted.

The paper closes with a reflective comparison of different approaches and the identification of future research challenges and recommendations. Extensive references to relevant literature appear throughout the survey, making this paper an excellent overview of the field and a perfect starting point for readers interested in exploring it in depth.