Activity Recognition in Pervasive Intelligent Environments

Activity recognition has attracted increasing attention as a number of related research areas such as pervasive computing, intelligent environments and robotics converge on this critical issue. It is also driven by growing real-world application needs in such areas as ambient assisted living and security surveillance. This chapter aims to provide an overview on existing approaches, current practices and future trends on activity recognition. It is intended to provide the necessary material to inform relevant research communities of the latest developments in this field in addition to providing a reference for researchers and system developers who are working towards the design and development of activity-based context aware applications. The chapter first reviews the existing approaches and algorithms that have been used for activity recognition in a number of related areas. It then describes the practice and lifecycle of the ontology-based approach to activity recognition that has recently been under vigorous investigation. Finally the chapter presents emerging research on activity recognition by outlining various issues and directions the field will take.

Providing cognitive assistance to Alzheimer’s patients in smart homes is a field of research that receives a lot of attention lately. The recognition of the patient’s behavior when he carries out some activities in a smart home is primordial in order to give adequate assistance at the opportune moment. To address this challenging issue, we present a formal activity recognition framework based on possibility theory and description logics. We present initial results from an implementation of this recognition approach in a smart home laboratory.

The advances of wearable sensors and wireless networks offer many opportunities to recognize human activities from sensor readings in pervasive computing. Existing work so far focus mainly on recognizing activities of a single user in a home environment. However, there are typically multiple inhabitants in a real home and they often perform activities together. In this paper, we investigate the problem of recognizing multi-user activities using wearable sensors in a home setting. We develop a multi-modal, wearable sensor platform to collect sensor data for multiple users, and study two temporal probabilistic models—Coupled Hidden Markov Model (CHMM) and Factorial Conditional Random Field (FCRF)—to model interacting processes in a sensor-based, multi-user scenario. We conduct a real-world trace collection done by two subjects over two weeks, and evaluate these two models through our experimental studies. Our experimental results show that we achieve an accuracy of 96.41% with CHMM and an accuracy of 87.93% with FCRF, respectively, for recognizing multi-user activities.

This paper introduces a framework for enabling context-aware behaviors in smart environment applications, with a special emphasis on smart homes and similar scenarios. In particular, an ontology-based architecture is described that allows system designers to specify non trivial situations the system must be able to detect on the basis of available sensory data. Relevant situations may include activities and events that could be prolonged over long periods of time. Therefore, the ontology encodes temporal operators that, once applied to sensory information, allow to efficiently recognize and correlate different human activities and other events whose temporal relationships are contextually important. Special emphasis is devoted to actual representation and recognition of temporally distributed situations. The proof of concept is validated through a thoroughly described example of system usage.

This paper describes the architecture and components of an AAL-enabling platform, centred around interactive TV (iTV), which combines OSGi middleware, RFID and NFC in order to ease the day to day of dependant or semi-dependant elderly people (its main focus), their care takers and relatives. The end result is an affordable, unobtrusive, evolvable, usable and easily deployable ICT infrastructure which aims to approach the vision of “AAL for All”. This is, it seeks a more widespread adoption of AAL and a better QoS on caretaking through the combination of common hardware, OSGi dynamic service and mobile-aided care data management.

An ontology-based context-aware framework for behavior analysis and reminder delivery is described within this Chapter. Such a framework may be used to assist elderly persons maintain a healthy daily routine and help them to live safely and independently within their own home for longer periods of time. Behavior analysis associated with the delivery of reminders offers strategies to promote a healthier lifestyle. Current studies addressing reminder based systems have focused largely on the delivery of prompts for a prescribed schedule at fixed times. This is not ideal given that such an approach does not consider what the user is doing and whether the reminder is relevant to them at that specific point in time. Our proposed solution is based upon high-level domain concept reasoning, to account for more complex scenarios. The solution, referred to as iMessenger, addresses the problem of efficient and appropriate delivery of feedback by combining context such as current activity, posture, location, time and personal schedule to manage any inconsistency between what the user is expected to do and what the user is actually doing. The ontology-based context-aware approach has the potential to integrate knowledge and data from different ontology-based repositories. Therefore, iMessenger can utilize a set of potential ontological, context extracting frameworks, to locate, monitor, address and deliver personalized behaviour related feedback, aiding people in the self-management of their well-being.

This paper deals with the smart house occupant prediction issue based on daily life activities. Based on data provided by non intrusive sensors and devices, our approach uses supervised learning technics to predict the house occupant. We applied Support Vector Machines (SVM) classifier to build a Behavior Classification Model (BCM) and learn the users’ habits when they perform activities for predicting and identifying the house occupant. To test the model, we have analyzed the early morning routine activity of six users at the DOMUS apartment and two users of the publicly available dataset of the Washington State University smart apartment tesbed. The results showed a high prediction precision and demonstrate that each user has his own manner to perform his morning activity, and can be easily identified by just learning his habits.

Although activity recognition is an active area of research no common benchmark for evaluating the performance of activity recognition methods exists. In this chapter we present the state of the art probabilistic models used in activity recognition and show their performance on several real world datasets. Our results can be used as a baseline for comparing the performance of other pattern recognition methods (both probabilistic and non-probabilistic). The datasets used in this chapter are made public, together with the source code of the probabilistic models used.

Humans deeply modified their relationship to their housings during the past centuries. Once a shelter where humans could find protection and have rest, the living place successfully evolved to become the midpoint of the family, the expression of own culture and nowadays a more self centered place where individuals develop their own personal aspirations and express their social position. With the introduction of communication technologies, humans may become nomads again with the ability to stay connected with others in any place at any time but, as a paradox, we can observe a wide movement for “cocooning”. Among all the services a living place can bring to inhabitants, we may list comfort, security, wellness and also health services. Thus a new living place is to be invented, becoming the “witness” of our breath, perceiving the inhabitants rhythms of activities, habits, tastes and wishes. Eventually, the “smart home” become the “Health Smart Home” to enable the follow up of physical and health status and meet the new concepts of “Aging in place” and “citizen health care”. We listed some of the research projects in Health Smart Home, which were launched worldwide to discover they are mostly based on very basic sensors and simple algorithms. We experienced our own Health Smart Home to prove that temporal analysis of data output from simple presence sensors is already worthwhile. We first produced “ambulatograms”, a temporal representation of the daily activity gathered from the presence sensors, and then discovered regular patterns of activities which we named “circadian activity rhythms (car)”, the direct relationship between night and day level of activities and also the information contained in periods of inactivity. We now concentrate on the automatic recognition of the daily Activities with multiple sensor fusions methods.

High-level (hierarchical) behaviour with long-term correlations is difficult to describe with first-order Markovian models like Hidden Markov models. We therefore discuss different approaches to synthesise generative probabilistic models for activity recognition based on different symbolic high-level description. Those descriptions of complex activities are compiled into robust generative models. The underlying assumptions for our work are (i) we need probabilistic models in robust activity recognition systems for the real world, (ii) those models should not necessarily rely on an extensive training phase and (iii) we should use available background knowledge to initialise them. We show how to construct such models based on different symbolic representations.

Activity and behaviour modelling are significant for activity recognition and personalized assistance, respectively, in smart home based assisted living. Ontology-based activity and behaviour modelling is able to leverage domain knowledge and heuristics to create Activities of Daily Living (ADL) and behaviour models with rich semantics. However, they suffer from incompleteness, inflexibility, and lack of adaptation. In this article, we propose a novel approach for learning and evolving activity and behaviour models. The approach uses predefined “seed” ADL ontologies to identify activities from sensor activation streams. Similarly, we provide predefined, but initially unpopulated behaviour ontologies to aid behaviour recognition. First, we develop algorithms that analyze logs of activity data to discover new activities as well as the conditions for evolving the seed ADL ontologies. Consequently, we provide an algorithm for learning and evolving behaviours (or life habits) from these logs. We illustrate our approach through scenarios. The first scenario shows how ADL models can be evolved to accommodate new ADL activities and peculiarities of individual smart home’s inhabitants. The second scenario describes how, subsequent to ADL learning and evolution, behaviours can be learned and evolved.

The automatic detection of complex human activities in daily life using distributed ambient and on-body sensors is still an open research challenge. A key issue is to construct scalable systems that can capture the large diversity and variety of human activities. Dynamic system reconfiguration is a possible solution to adaptively focus on the current scene and thus reduce recognition complexity. In this work, we evaluate potential energy savings and performance gains of dynamic reconfiguration in a case study using 28 sensors recording 78 activities performed within four settings. Our results show that reconfiguration improves recognition performance by up to 11.48 %, while reducing energy consumption when turning off unneeded sensors by 74.8 %. The granularity of reconfiguration trades off recognition performance for energy savings.

As the average age of the population increases worldwide, automated tools for remote monitoring of activity are increasingly necessary and valuable. This chapter highlights embedded systems for activity recognition that provide privacy, do not require major infrastructure,and are easy to configure. The strengths and weaknesses of popular sensing modes that include RFID, motion, pressure, acceleration, and machine vision are discussed. A new activity detection system is also described for high privacy area like the bathroomand bedroom environment.

Obesity is an increasing problem for modern societies, which implies enormous financial burdens for public health-care systems. There is growing evidence that a lack of cooking and food preparation skills is a substantial barrier to healthier eating for a significant proportion of the population. We present the basis for a technological approach to promoting healthier eating by encouraging people to cook more often. We integrated tri-axial acceleration sensors into kitchen utensils (knifes, scoops, spoons), which allows us to continuously monitor the activities people perform while acting in the kitchen. A recognition framework is described, which discriminates ten typical kitchen activities. It is based on a sliding-window procedure that extracts statistical features for contiguous portions of the sensor data. These frames are fed into a Gaussian mixture density classifier, which provides recognition hypotheses in real-time. We evaluated the activity recognition system by means of practical experiments of unconstrained food preparation. The system achieves classification accuracy of ca. 90% for a dataset that covers 20 persons’ cooking activities.