Ebook: Behaviour Monitoring and Interpretation – BMI

The workshop on Behaviour Monitoring and Interpretation (BMI) was launched in 2007 and was co-located with the German conference on Artificial Intelligence. The first two editions of the workshop indicated a significant interest in research related to BMI and motivated the creation of the current volume with contributions by a number of leading researchers in this emerging field. Although the book covers a broad spectrum of topics concerned with behaviour monitoring and interpretation, there are essentially two prominent directions in focus here based on their particular interest in the ongoing research: the investigation of motion patterns and the area of Ambient Assisted Living. This volume aims to offer state-of-the-art contributions on these directions of research.

The first chapter is an introduction to the area of BMI by Björn Gottfried and Hamid Aghajan; it explains what this field signifies and how it relates to other research areas. Then, in the first part of this volume, a number of chapters discuss recent developments in monitoring and representing behaviour, with a particular focus on movementbased behaviour. Alexandra Millonig, Norbert Brändle, Markus Ray, Dietmar Bauer, and Stefan Van Der Spek provide an overview of methods for monitoring and analyzing pedestrian motion behaviours. The subsequent chapter by Patrick Laube also considers movement behaviour; however, the focus is on which typical patterns can be distinguished that are not restricted to human beings and also involve patterns of groups of objects. Similarly, groups and their movement patterns are investigated by Zena Wood and Antony Galton who provide a classification scheme for collectives. Yohei Kurata and Max Egenhofer provide a qualitative spatial representation for relating directed line segments to their topological context; in this way they characterise movement patterns of individuals in relation to their topologically described context. Another qualitative representation is provided by Frank Dylla who considers ordinal relations; basic movement patterns are described between pairs of objects.

The second part of the volume includes chapters that are more application driven. Tim Adlam, Bruce Carey-Smith, Nina Evans, Roger Orpwood, Jennifer Boger, and Alex Mihailidis present case studies about the monitoring and support of people with dementia in smart environments. Sylvain Giroux, Tatjana Leblanc, Abdenour Bouzouane, Bruno Bouchard, Hélène Pigot, and Jérémy Bauchet report on AI techniques applied in smart environments, in particular for providing inhabitants with cognitive impairment assistance in their everyday life. Joyca Lacroix, Yasmin Aghajan, and Aart Van Halteren discuss another approach that makes environments smarter: ambient assisted physical activity systems are presented that aid in increasing the engagement of seniors in physical activities.

The third part presents a number of investigations which show how monitored behaviours can be interpreted in smart environments. Peter Kiefer, Klaus Stein, and Christoph Schlieder give a survey on knowledge-intensive methods for intention recognition; in particular, they look at how environments are spatially structured and take into account context-specific background knowledge. Albert Hein, Christoph Burghardt, Martin Giersich, and Thomas Kirste discuss an approach for the detection of high-level activities, in particular for interpreting team behaviours. Asier Aztiria, Alberto Izaguirre, Rosa Basagoiti, and Juan Carlos Augusto present a model for how ambient intelligence systems can automatically discover patterns of user behaviour; they also discuss how the interaction of the users with the system can improve the performance of their system.

The two final chapters are devoted to the infrastructure of smart environments. Matt Duckham and Rohan Bennett investigate decentralised spatiotemporal algorithms that optimise the support of spatially distributed systems of smart environments; for example, in order to monitor environmental changes even at large geographical scale. More related to middleware technologies is the contribution by Alvaro Marco, Roberto Casas, Gerald Bauer, Rubén Blasco, Ángel Asensio, Bruno Jean-Bart, and Miriam Ibanez; they present a framework for enabling the interoperability and handling the heterogeneity of components found in ambient assisted living systems.

We hope you will find the presented material in this volume of interest to your research. A further motivation for introducing this book has been to encourage interdisciplinary interaction among researchers working on the various fields related to BMI. We hope the presented state-of-the-art in this volume will offer a glimpse of the potentials ahead.

June 2009, Björn Gottfried and Hamid Aghajan

This chapter provides an overview of the area of Behaviour Monitoring and Interpretation, BMI for short. It outlines this research direction and gives examples of current research. In a nutshell, BMI is about monitoring behaviour of humans or other entities and interpretation of the observed behaviour. Then, the chapter shortly discusses how BMI compares to related areas such as Ambient Intelligence and ubiquitous computing. Some details about future challenges are finally shown.

The investigation of pedestrian spatio-temporal behaviour is of particular interest in many different research fields. Disciplines like travel behaviour research and tourism research, social sciences, artificial intelligence, geoinformation and many others have approached this subject from different perspectives. Depending on the particular research questions, various methods of data collection and analysis have been developed and applied in order to gain insight into specific aspects of human motion behaviour and the determinants influencing spatial activities. In this contribution, we provide a general overview about most commonly used methods for monitoring and analysing human spatio-temporal behaviour. After discussing frequently used empirical methods of data collection and emphasising related advantages and limitations, we present seven case studies concerning the collection and analysis of human motion behaviour following different purposes.

This chapter reports on progress made in the development of techniques and tools to formalize, describe, detect and understand movement patterns in spatiotemporal data. The chapter is presented as a critical review of the relevant literature in the fields of geographical information science, data mining and knowledge discovery, and computational geometry. After discussing the nature of movement patterns, several conceptual space models accommodating movement are discussed as a preliminary for the formation of movement patterns. Then typical types of movement patterns and their signature applications are portrayed. The chapter concludes with a series of limitations to movement pattern analysis, an outlook, and a research agenda for future work.

Behavioural monitoring often concerns the interpretation of the motion of an agent with respect to an area of interest. Geometrically, the trajectory of the agent is represented by a directed line segment (DLine) around/over a region. Topological relations between a DLine and a region concern how the DLine intersect with the region and, therefore, these relations are useful for characterizing the motion in association with an area of interest. In this chapter, we introduce a formal model of topological DLine-region relations and the application of these relations for the characterization of motions. We start from a model of topological relations between a non-directed line and a region, called the 9-intersection, reviewing how these line-region relations are associated with spatial predicates. Then, we introduce the 9⁺-intersection, which distinguishes 26 topological DLine-region relations in , with which we explore several approaches to associating motions with the model of human motion concepts. Finally, we introduce two future research questions: (i) simplification of complex trajectories by segmentation and (ii) mapping between non-planar motion concepts and topological DLine-region relations in .

In this chapter we show how qualitative representations can be applied for the formalization of agent behavior. First, we give an introduction to several aspects of space relevant for agent motion and agent control, especially orientation, location, and distance. Based on these preliminaries we explain which characteristics a qualitative spatial representation and its operations, called qualitative calculus, should have so that the knowledge can be manipulated in an adequate manner. Afterwards, we present the two main categories of reasoning with qualitative calculi: constraint-based reasoning and neighborhood-based reasoning, action-augmented neighborhood respectively. Exemplary, we sketch how this structure can be utilized for representing agent behavior. Additionally, we apply the approaches to agent control in the context of right-of-way rules in sea navigation.

Collective phenomena and their associated movement patterns are ubiquitous in everyday life. However, formal reasoning about these phenomena is currently hampered by the lack of adequate tools. We have previously developed a classification of collectives but this is incomplete with regards to movement and therefore needs to be integrated with a classification of collective motion. This paper analyses existing research into the movement patterns of collectives. Although there has been some research into the movement patterns of particular kinds of collectives, most of this is found to focus on the level of the individuals; vital information about the collective is lost. Existing research that focuses on movement patterns improves on this but still leaves many questions unanswered and important features of collectives unable to be represented. Therefore, we develop a list of goals which we believe a classification of collective motion needs to satisfy, and introduce the foundations of a system that we believe will satisfy these goals. We hope that this work will provide a sound basis for the development and formalisation of a comprehensive classification of collectives and their motions.

This chapter is about the application of behaviour monitoring technology in the context of a smart home for people with dementia. It is not about the design of technology, but about the application and configuration of existing technology in a specific context: in this case, smart flats for people with dementia in London and Bristol. Technology was installed and evaluated in a year long evaluation in London by a resident tenant. He was assessed throughout his tenancy using standardized outcome measures, by clinical professionals; and through the analysis of data collected by sensors installed in his flat. It was demonstrated that the technology had a positive impact on his life, improving his sleep in particular. This improvement had a positive effect on many other aspects of his life in the extra care setting where he lived. The Bristol evaluation is in progress. It is also an evaluation of smart home technology embedded in a person's own home.

This chapter also describes two technologies being developed at the University of Toronto, in Canada. The first is COACH, a system used for the guidance of activities of daily living, and the second is HELPER, a fall detection and personal emergency response system (PERS). These technologies operate autonomously with little or no explicit input from the person using them, making them extremely intuitive and effortless to use. Practical experience and clinical results gained from the latest efficacy trials with COACH are presented and discussed. From the data collected through these trials, it seems that COACH has a positive effect on peoples' ability to independently complete the activity of handwashing. It is hoped that monitoring technologies such as these will improve the independence and quality of life for people with dementia.

The current and prospective situation of cognitively impaired people entails great human, social, and economical costs. Smart homes can help to maintain at home cognitively impaired people, to improve their autonomy, and accordingly to alleviate the burden put on informal and professional caregivers. This chapter will provide a comprehensive view of the research performed at DOMUS lab. This research aims at turning the whole home into a cognitive prosthetic, especially by providing cognitive assistance. The first part of the chapter presents research on the infrastructure, both sensors networks and middleware. Research work on autonomic computing, multi-person localization, context awareness, and personalization are presented. The next part of the chapter illustrates by means of DOMUS research prototypes how cognitive assistance can help to address four kinds of cognitive deficits: initiation, attention, planning, and memory. Studies involving cognitively impaired people are also be presented. In the final part of the chapter, the role of AI, context awareness and behavior tracking are questioned. To what extend are they compulsory? Does design can provide smart and simple solutions to complex issues?

Although many people are aware of the beneficial effects of physical exercise on health, they often experience difficulties in finding time and motivation to incorporate regular gym visits into their busy schedules. Exercising at home is a viable alternative, but lacks the human coaching and social support that is available in traditional exercise settings. Modern technology opens an enormous space of possibilities for enriching the in-home exercise experience with motivational coaching and social factors through interactive monitoring, feedback, and social connectedness. In this chapter we introduce an interactive social in-home exercise system, which enables the monitoring and sharing of exercise movements through an avatar in a social context (composed of a coach or other exercisers). We conduct a survey study in two user groups to explore to what extent users appreciate the use of avatars to visualize their exercise movements and to gain insight into preferences with respect to the level of personalization of avatars in various social contexts. Also, we examine the relationship between several user characteristics (body image, gender, and age) and these avatar preferences. The results show that individuals overall like the use of avatars to visualize exercise movements. Moreover, we find that the appreciation of visualizing exercise movements through an avatar and the preferences with respect to the level of personalization vary with the type of social context, with a greater appreciation for the avatar when exercising in private than when the avatar is communicated in a social context of unknown others. Also, in one group we find a greater appreciation to share the avatar with a coach than with friends or with unknown others. Finally, the results suggest that body image, gender, and age play an important role for avatar preferences.

A central design issue in ambient assisted living consists in creating environments which show a smart behavior that is transparent to the user in the sense that the effects of ambient intelligence are easily predictable. Another aspect of behavioral transparency is related to the effective and efficient communication with the user about relevant background knowledge. Rule-based specifications are widely used in today's home automation systems as a means to communicate user knowledge.

This chapter addresses ambient intelligence based on the recognition of the user's intentions to act and discusses approaches that identify user intentions in the context of specific background knowledge about possible tasks and the spatial environment. We give a survey on knowledge-based methods for intention recognition and compare different rule-based formalisms with regard to the trade-off between expressiveness and complexity. Special emphasis is laid on approaches that assume a user moving in an environment which is spatially structured by a partonomy as most indoor and near outdoor environments are.

Situation awareness is a critical prerequisite for providing proactive assistance in the context of Ubiquitous Computing. In this chapter we will focus on the detection of high-level team intentions. At first we introduce an exemplary scenario of a team meeting in a smart environment and outline relevant criteria for our approach. Then we give a short insight into the current state of the art, discussing related research projects while trying to find points of intersections for a more generic modeling approach. We reflect team behavior and individual problem solving strategies against the background of social and cognitive psychology and review sources of prior information like domain knowledge and common sense to outline a basic generic model structure. We will argue that especially probabilistic model based approaches are capable of efficiently and robustly inferring activities by making use of this knowledge. We will take a closer look at models which are able to represent human behavior at different levels of abstraction, accompanied by a short explanation of their reasoning mechanisms for inference and prediction. At last we take a look at the synthesis of such models using planning methods for behavior sequences.

Intelligent Environments are supposed to act proactively anticipating the user's needs and preferences in order to provide effective support. Therefore, the capability of an Intelligent Environment to learn user's habits and common behaviors becomes an important step towards allowing an environment to provide such personalized services. In this chapter we explain and exemplify the importance of detecting patterns of behavior, we propose a system which learns patterns of behavior and also a complementary interaction system, based on speech recognition, which facilitates the use of such patterns in real applications.

Computing in distributed systems increasingly occurs somewhere, with that location being integral to computational process itself. Ambient spatial intelligence (AmSI) is concerned with embedding the intelligence to respond to spatiotemporal queries and monitor geographical events in built and natural environments. The emergence of AmSI is enabled by new spatial computing technology: geosensor networks (wireless networks of sensor-enabled, location-aware computers monitoring environmental change). This chapter argues that decentralized spatial computing (DeSC), where spatial information is partially or completely filtered, summarized, or analysed in a geosensor network, is a fundamental technique required to support AmSI. By contrast, existing models for complex spatiotemporal analysis and queries almost always adopt a centralized approach to computation, where global spatial data is collated and processed, for example in a spatial database or GIS. The chapter identifies four main research challenges facing DeSC (dealing with uncertainty, dynamism, information integration, and interfaces), and illustrates the importance of DeSC to AmSI with reference to three key AmSI applications: management of the environment, infrastructure, and people. Finally, a legislative analysis of one specific domain of human activity, government acts in Australia, illustrates the potential for increasing importance of AmSI in the near future.

Ambient Assisted Living (AAL) is currently being studied by research institutions and industry, and it is expected to become a reality in the near future. One of the main issues in AAL is the heterogeneity of technology, which demands a major effort to allow interoperability of elements operating in a smart environment. Service-oriented frameworks such as OSGi provide beside other interesting facilities an adequate way to handle this heterogeneity and are often used as framework platforms which support smart home environment systems. However, there is also heterogeneity among different AAL projects, as every research team develops its own smart environment. Thus, devices and services definitions usually differ from one project to another. In the following a proposal for a Common OSGi Interface is introduced, which publishes both interfaces for devices and services of AAL related scenarios. By using these interfaces, an AAL application designer will be able to take advantage of existing device and service developments and ensure interoperability with other systems.