A survey on ontologies for human behavior recognition
ACM Computing Surveys (CSUR)
| Hi-index | 0.00 |
With the advance and prevalence of low-cost, low-power computing devices it is becoming increasingly clear that we are going to work and live in intelligent environments in the very near future. An intelligent environment is a physical space, e.g. an office, a hospital, a home or a car, wherein miniature computing devices perceive, monitor, and interact with the environments and human users within them. This insight and vision has provoked considerable research leading to a number of emerging research areas, such as mobile computing, ubiquitous/pervasive computing, sensor network and ambient intelligence. There has also been substantial progress on the conceptualization and design of innovative application scenarios, technological infrastructure and system prototypes. For example, a compelling real-world intelligent environment is "Smart Homes" -- an augmented home environment within which the daily activities of its inhabitants, usually the elderly or disabled, are monitored and analysed so that personalized context-aware assistances can be provided. A central research issue about the applications of intelligent environments is activity recognition, i.e. to identify what users perform based on the sensor observations of the environment. Only when users' activities are recognised, can application systems or devices support the users to accomplish their intended activities in an optimal way, e.g. to predict next actions, to suggest required resources, to explain the situation or to help users by taking actions through actuators. As such, activity recognition has attracted increasing attention from a diversity of research communities, and numerous methods have been studied and experimented, notably the data-centred data mining approach using probabilistic analysis and machine learning. This tutorial aims to introduce an ontology-based approach to activity recognition. It will first have a brief look at current methods and practices on activity recognition. Then it will analyse the nature and characteristics of intelligent environments and user behaviours. Based on this analysis the tutorial will argue the necessity of an ontology based approach, in particular, the benefits of exploiting domain knowledge for activity modelling and reasoning. Following this we shall discuss the core components and enabling technologies that underpin the approach, and a system architecture for its realisation. To demonstrate the approach, the tutorial will be accompanied with examples and screenshots from our work on modelling and assisting activities of daily living in Smart Homes.