Recognising Human Behaviour in a Spatio-Temporal Context

Recognising Human Behaviour in a Spatio-Temporal Context

Hans W. Guesgen (Massey University, New Zealand) and Stephen Marsland (Massey University, New Zealand)
DOI: 10.4018/978-1-61692-857-5.ch022


Identifying human behaviours in smart homes from sensor observations is an important research problem. The addition of contextual information about environmental circumstances and prior activities, as well as spatial and temporal data, can assist in both recognising particular behaviours and detecting abnormalities in these behaviours. In this chapter, we describe a novel method of representing this data and discuss a wide variety of possible implementation strategies.
Chapter Preview

An Ai Perspective On The Smart Home

The smart home can be separated into the sensory system and the ambient intelligence that works to interpret the sensor observations. In this chapter, we do not explicitly consider what types of sensor are (or could be) available in the home, but assume that the sensory stream is available in the form of a sequence of ‘tokens’, i.e., there has been some preprocessing of the sensor readings (and possibly the fusion of different sensor data) into a sequence of observations that can be used by some form of ambient intelligence system. For our purposes, we consider that the task of the smart home is to segment this token stream into different behaviours, and to identify whether or not this behaviour is typical of the user and, if not, whether it is sufficiently abnormal to warrant any action (such as calling a carer, or interacting with the inhabitant, for example in the form of reminders: ‘did you remember to turn the gas off?’).

Our interpretation of the smart home problem from the point of view of artificial intelligence is that it requires solutions to at least some of the following problems:

Key Terms in this Chapter

Neural Network: Computational model inspired by how neurons work in the brain.

Temporal Calculus: Logic-based approach for reasoning about time.

Hidden Markov Model: Probabilistic graphical model that uses a set of hidden (unknown) states to classify a sequence of observations over time.

Activity of Daily Living: Set of activities performed with a particular aim in mind, often reoccurring on a regular basis.

Novelty Detection: Recognising behaviours that are new or unusual.

Behaviour Recognition: Identifying and classifying activities on the basis of observations such as sensor data streams.

Sensor Observation: Data stream consisting of tokens associated with sensor readings.

Lifelong Learning: The ability of the system to adapt to changes in the environment and in human behaviours.

Spatial Calculus: Logic-based approach for reasoning about space.

Complete Chapter List

Search this Book: