Ebook: State of the Art in AI Applied to Ambient Intelligence

Imagine a future where human environments respond to human preferences and needs. In this world, smart devices equipped with intelligent features and the capability to sense, communicate, and support humans in daily life activities will be unremarkable features. During the last decade, questions were raised on how intelligent were the so called “smart systems”or “Ambient intelligence environments”. Nowadays, the word smart has become a sort of wild card to be attached to any new product introduced to the market.

The research area related to Smart Homes has blended with other areas in Computer Science and expanded in several directions. Popularity has grown exponentially and several fields of applications are now closer to reality rather than science fiction vignettes. There are Smart Cars, Smart Classrooms, Smart Farming and many other smart environments in which technology is changing the way people relate to them. Thus, we will expect cars to warn us of hazards, track our location and provide timely route advice. We will speak to simple machines and hold conversations with more complex systems, such as intelligent homes that will help us monitoring conditions, tracking routine tasks, and programming the behaviour of the heater, lights, garden watering and the entertainment centre.

It was anticipated a decade ago that the analysis of the extent and the way that Artificial Intelligence (AI) can benefit Ambient Intelligence was not obvious, although it was clear that AI seemed in a good position to contribute to this new emerging field. There are still many interesting open questionsfor example“What is good use of AI in Ambient Intelligence?”, because there are ethical dimensions to this, and “How much intelligence is needed to really make a difference?”, or “How do people expect to find intelligence in their surroundings?”.

This book aims at providing a clear picture of what has been achieved after a decade of discussion and, most importantly, what avenues are the most promising for exploration in the next decade. Indeed, this volume aims at creating a reference that field experts may use to be updated and inspired to create better systems for society.

In the chapter “A survey on applying machine learning techniques for behavioural awareness”, the focus is on reviewing the state-of-the-art of the applicability of machine learning techniques for behavioural awareness, considering both individuals and group behavioural recognition.

The chapter “Modelling spatial and temporal context to support activity recognition” introduces the advances in context awareness with the dawn of mobile computing and the Internet of Things. In this sense, new areas such as affective computing and biometrics are emerging as new opportunities for context awareness. Then, the chapter entitled “Affect aware ambient intelligence: current and future directions” provides a review of current research aiming to offer an insight for affective computing by exploring all key aspects relating to development of an affective human-centered system. The chapter entitled “Behavioural biometrics and ambient intelligence: new opportunities for context-aware applications” describes a multi-faceted smart environment for the acquisition of relevant contextual information about users, such as performance, attention, mental fatigue and stress.

Following chapters focus on application of AI to real Ambient Intelligence environments. The chapter “Energy and environmental long-term monitoring system for inhabitants well-being” presents a monitoring system capable of measuring the energy consumed by end-users and a set of environmental parameters and, then, influencing user comfort developing social interaction algorithms. Next two chapters entitled “Behavioural patterns from cellular data streams and outdoor lighting as strong allies for smart urban ecosystems” and “Learning daily routines in smart office environments” present two use cases where AI techniques are used to simplify daily activities of users in two different environments, i.e., respectively, smart cities and smart offices.

Finally, the last two chapters show some of the necessary technological tools that are necessary to develop ambient intelligence environments. In “ECKRUCAmI architecture – applications in healthcare domain” the authors propose an architecture dedicated to a specific domain, which aims to facilitate the development of more efficient healthcare supporting environments. The chapter “A qualitative image descriptor QIDL⁺N to obtain logics and narratives applied to ambient intelligence systems” presents a model for obtaining a real-world scene description.

This book is part of the outcome of the Workshop on Artificial Intelligence Techniques for Ambient Intelligence (AITAmI) which has run for 10 consecutive editions. The event was created as a way to investigate some of the many open questions on the intelligence features that a place requires to realize the concept of Ambient Intelligence. During a decade we have explored a number of those questions and challenges and we have contributed to progress towards a better understanding of the overall problem. This book is representative of the diversity of approaches and issues which are currently being considered.

Editors: Asier Aztiria, Juan Carlos Augusto, Andrea Orlandini

This survey reviews the state-of-the-art on the applicability of machine learning techniques for behavioral awareness. Not only is behavioral awareness an important capability of ambient assisted living environments to better support individuals during their activities of daily living, it is also gaining attention as a way to identify individuals based on distinguishing or unique features in their behavior. In this work, we not only pay attention to individual behavior but also to behavioral traits of a group of individuals as a whole. This survey documents a broad overview of the state-of-the-art in behavioral recognition, learning and mining. It provides an overview of how activities can be recognized through the identification and classification of certain features in event-based data streams. This overview highlights a variety of well-known probabilistic learning and prediction methods and how they have been applied for the purpose of identification and activity recognition.

With the dawn of mobile computing and the Internet of Things, context-aware computing has gained increasing attention. The idea is to not only consider explicit inputs when computing an output, but also to consider contextual factors such as location, time, and environmental data. Context-aware computing is not restricted to mobile computing and the Internet of Things, but is also beneficial in other areas. The area we focus on in this chapter is activity recognition in smart environments.

In the last ten years or so, a number of approaches have been developed that aim to recognise activities in smart environments such as smart homes or smart offices, ranging from logic-based approaches to probabilistic machine learning approaches. These approaches still have deficits and only work under certain assumptions. We argue in this chapter that using context information can lead to an improvement in activity recognition. To prove our point, we focus on spatial and temporal context information and introduce a number of methods to reason about this type of information.

The promise of a reality where intelligent technologies and unobtrusive networks of interconnected electronic devices support individuals in their daily tasks and activities is delivered by the vision of Ambient Intelligence. In order for this vision to be realized modern technology should posse an understanding of a core aspect of human nature, namely emotion. Affective Computing is a multidisciplinary scientific field which has emerged in order to deliver applications which are able to recognize their user's emotion, produce affect sensitive outputs in order to aid them and respond in a human like manner, resulting in a higher level of human machine interaction. This chapter provides a review of current research aiming to offer an insight into affective computing by exploring all key aspects relating to the development of an effective human centered system and by identifying open questions and future research directions for this scientific area.

Ambient Intelligence has always been associated with the promise of exciting new applications, aware of the users' needs and state, and proactive towards their goals. However, the acquisition of the necessary information for supporting such high-level learning and decision-making processes is not always straightforward. In this chapter we describe a multi-faceted smart environment for the acquisition of relevant contextual information about its users. This information, acquired transparently through the technological devices in the environment, supports the building of high-level knowledge about the users, including a quantification of aspects such as performance, attention, mental fatigue and stress. The environment described is particularly suited for milieus such as workplaces and classrooms, in which this kind of information may be very important for the effective management of human resources, with advantages for organizations and individuals alike.

The increasing demand for building services and comfort levels, together with the increased time spent inside buildings, assures an upward trend in long-term monitoring system demand for the future. In this paper, we present the work done for designing added value human well-being services starting from a state-of-the-art continuous data gathering infrastructure. The paper presents the proposed energy and environmental long-term monitoring system that is able to measure both the energy consumed by end users and the environmental parameters in office environments. The paper shortly describes the general idea, then it focuses on the work done to create a well-being service on the top of the data gathering layer. In particular, it dwells on the deployment approach focusing on the description of the long-term monitoring system and providing preliminary results of the proposed real-time social interactions algorithm.

The concept of smart urban systems assumes tracking and analysing various signals generated in the environment surrounding inhabitants. The aim is to provide responses which are context-aware and pro-active and allow to increase the inhabitant's life comfort. Smart outdoor lighting solutions are common and mandatory components for modern urban spaces and municipalities offering considerably energy savings. On the other hand, inhabitants' behavioural fingerprints originating from cellular data streams, produced and stored in base transceiver stations (BTSs), are the most ‘democratic’ and ubiquitous information which might support modern solutions tuning/fitting lighting parameters to the current needs. This paper presents the fundamental problems of elicitation, classification and understanding of such signals/data for the development of smart systems operating in urban areas with the special focus put on smart outdoor lighting problems. Omnipresence of computing is strongly focused on providing on-line support to users/inhabitants of smart cities. Three important components underlying computational model are discussed in this paper: the method of analysing selected elements of mobile phone datasets through understanding inhabitants' behavioural fingerprints, the multi-agent system supporting the proposed logic and the formalism based on graphs that allows reasoning about inhabitant behaviours. Corresponding scenarios for public transport and outdoor lighting are outlined as well.

One of the goals of smart environments is satisfying proactively its users' needs. To do so, it should understand what they are doing and what are their habits and preferences. In turn, this requires recognizing, at a lower level, single activities (e.g., reading, sitting, looking for books, …) and, at a higher one, how these activities are organized into daily routines. For these purposes we exploit the WoMan system. Using a process mining approach, it is able to incrementally learn a user's activities and daily routines as workflow models. Then, such models are exploited by WoGue (“Workflow Guesser”) system to recognize the current routine or activity of the user. The approach has been tested in a real-world setting, a smart office environment, equipped with a sensor network based on Arduino. We collected an annotated dataset of 45 days, and learned from it the workflow models of the user's daily routines and of his activities performed in the office. Results of some experiments show that how our approach is quite effective to learn and recognize activities and routines. Indeed, it achieves an average accuracy of 82% for activities and 98% for transitions among activities. Also the real-time recognition performance was tested, using sensor data coming from the smart office environment. In 82% of the cases the system correctly recognized the current activity and routine after just four events.

Ambient Intelligence is a paradigm in Artificial Intelligence which promotes intelligent computing and responds to the necessities of every person in many different environments whether it is a home, classroom, a shopping, or a hospital. Nowadays we are increasingly seeing and talking about issues such as aging populations or the increase in the number of patients with chronic diseases (among other issues). Looking at the solutions available in this area it is necessary to develop more effective systems that can support people and improve their quality of life. This work is a brief study on EKRUCAmI architecture and its suitability to the healthcare sector. Some healthcare systems will be presented which have been designed to use EKRUCAmI architecture.

A model for obtaining a real-world scene narrative using qualitative features of shape, colour, topology, location and size is presented in this paper (QIDL⁺N). Its main aim is describing the location of target objects with respect to known or unknown objects in a scene. The QIDL⁺N produces: (i) a logical description in Prolog for users to solve queries on the knowledge base built for each scene or for agents to reason about that scene; and (ii) a narrative in English language to read or listen to which may improve human-AmI-systems interaction, specially helping those users with special needs. A use case of a top desk scene is used to illustrate the approach and promising results are obtained.