Multi-user activity recognition: Challenges and opportunities

Human activity recognition has attracted enormous research interest thanks to its fundamental importance in several domains spanning from health-care to security, safety, and entertainment. Robust and consolidated literature focused on the study of activities performed by single individuals, with a great variety of approaches in terms of sensing modalities, recognition techniques, a specific set of recognized activities, and final application objectives. However, much less research attention has been devoted to scenarios in which multiple people perform individual or joint actions and activities forming groups to achieve given common goals. This problem is often referred to as multi-user activity recognition. With the advent of the Internet-of-Things, smart objects are being pervasively spread in the environment and worn on the human body, enabling contextual and distributed recognition of group and multi-user activities. Therefore, this survey discusses clear motivations and advantages of multi-user activity recognition based on sensing methods, recognition approaches, and practical applications with attention to related data fusion challenges and techniques. By identifying the critical aspects of this multi-faceted problem, the survey aims to provide a systematic categorization and comparison framework of the state-of-the-art that drives the discussion to important open research challenges and future directions.