Combining Timed Data and Expert's Knowledge to Model Human Behavior

Abstract

One of the major issues of monitoring activities in smart environments is the building of activity models from sensor's timed data. This work proposes a general theoretical approach to this aim, based on a Knowledge Engineering methodology and a Machine Learning process that are both funded on a general theory of dynamic process modeling, the Timed Observation Theory. In the proposed framework, activity recognition is an abstraction process where the activities are conceived as entities at di erent abstraction levels. This paper aims at showing that prior expert's knowledge about resident activities can be compared with posterior knowledge induced from timed data. The proposed approach is described through the database of the prototypical home of the GerHome project.

Keywords

• eol>Smart Environment
• Human Activity
• Dynamic Process Modeling
• Machine Learning

-

tivities of daily living (IADL) increase with the age and, in general, are higher among people 60 years or more [ 29 ]. These restrictions a ect the autonomy and the well-being of people in the last phases of life. Many older adults or people with disabilities wish to remain in their home for as long as possible even when their daily needs are a ected. Certain surveys indicates that nearly 75 percent of respondents age 45 or older hope to stay in their homes as they age [ 1 ]. Under these considerations, and taking into account that the autonomy of a person depends not only of its capacities to accomplish acts of daily life, but also on the possibilities that the environment can provide, there is a growing interest in observing ADLs and monitoring health through smart environments [ 33, 6 ].

A smart environment "is able to acquire and apply knowledge about an environment and also to adapt to its inhabitants in order to improve their experience in that environment" [ 8 ]. An example of smart environment is a smart home as Aware Home [ 2 ], EasyLiving [ 5 ], MavHome [ 7, 17 ], CUS Smart Home [ 31 ], iDorm [ 16 ], QuoVADis [ 15 ], CASAS [ 28, 24 ] and GerHome [ 34, 35 ], where inhabitant behavior is recorded by sensors and monitored by a program in order to detect the activity carried out (such as cooking, eating, watching TV, etc.).

Activity monitoring consists of comparing resident behavior with activity models to determine the executed activity and to detect anomalies or behaviors that require automatic intervention of the environment. Nevertheless, the de nition of models for human activity monitoring is one of the major issues due to the randomness of human behavior and, therefore, to the subjective notion of the concept activity. The work presented in this paper proposes a general theoretical framework which conceptually de nes the notion of activity and relates Knowledge Engineering methodologies with Data Mining techniques to de ne and to identify resident activities. The application of this approach is illustrated through the GerHome project of Centre Scienti que et Technique du Ba^timent (CSTB, France). The aim of this project is to develop technical solutions to the problem of providing greater autonomy and better quality of life to the elderly at home; and thus, to work on the prevention of accidents such as fall down originating in the frailty increase of the person. The hold method is to track the frailty trends by monitoring the daily activity and to compare a learned model with sensor data recorded from e ective activities. This research path leads the way to detect activity models and could o er an appropriate and reliable method to extract relevant and coherent daily activity patterns. In Section 2, we introduce related works and the motivation of our approach. Section 3 presents the theoretical framework proposed for modeling and recognizing the resident's activities. Section 4 describes our proposal applied to the GerHome project. Finally, in Section 5, our conclusions are presented.

2. RELATED WORKS

Human activity recognition in perceptual environments involves severe challenges due to the erratic nature of human behavior. To determine what is being done can be complicated if di erent activities are executed at the same time; e.g., to cook while watching TV. Besides, the same detected action can be associated with several activities depending on the context in which it is carried out then, to discriminate what is the right activity is not trivial; e.g., to open sink water tap can be part of cooking or washing dishes. Moreover, activities can be interleaved: while washing dishes the phone rings, the activity is paused, the phone is answered and then, the activity is taken up again. Thus, to determine what a person is doing at a particular time is not a simple task.

The problem lies in the meaning and the interpretation of the perceptual inputs due to the large gap that exists between the low level signals, as pixels, sensor signals, etc., and that one that is inferred in a higher level, for example, washing dishes.

Di erent works propose a characterization and a de nition of human activity in smart environments. In particular, an activity can be considered in terms of space (activity location), of time (temporal patterns), of goals (intentions) and in terms of ethnographic data [ 3 ]. On the other hand, as proposed in [ 20 ], activities are directly linked with human acts which can be speci ed by constructing a probabilistic context-free grammar (PCFG), whose alphabet consists of poses (Figure 1(a)). Thus, activity recognition is based on a successive abstraction process where human activities are de ned from the visual observation of body poses obtained from video data and, three levels of abstraction are conceived (Figure 1): continuous signal (optical ow), discrete event (body pose) and activity (sequence of discrete events). In [ 19 ], activities depend on temporal, logical and causal constraints linked with an intention and three abstraction levels are also presented: low level sensory stim(a) Activity de nition (Pick Up) [ 18 ] (b) Motion patterns from sensor data (Sit and Stand ) [ 20 ] uli, notion of causality amongst some qualitative activity descriptors and notion of context-sensitive intent. Similarly, [ 21 ] proposes three levels of abstraction as well: movements as low-level semantic primitives, activities as sequences of states and movements and human behavioral actions as high level semantic events.

The MavHome (managing and adaptive versatile home) project [ 32 ] is focused on providing smart environments, whose goals are to maximize the comfort of the inhabitants, minimize the consumption of resources, and maintain the safety of the environment and its residents. In this project, once again, three levels of abstraction are proposed (discrete events coming from sensors, event sequence and activity) and the move from an abstraction level to the other is based on models that are produced using a process of Knowledge Discovering from Databases (the Apriori algorithm [ 30 ] or Hiden Markov models, Figure 2). A similar approach is used in CASAS [ 27, 25, 26, 23 ], an adaptive smart home system that discovers and adapts to changes in the resident's preferences in order to generate satisfactory automation policies. In this case, a temporal point of view about the di erent levels of abstraction (Figure 3) is considered. It is to note that CASAS uses an algorithm of pattern adaptation miner (PAM) in order to adapt to changes in the behavior patterns.

All of these approaches include the idea of hierarchical abstraction, and de ne three levels of abstraction (Table 1): discrete events, sequence of discrete events and a taxonomic classi cation at the highest level. Nevertheless, these definitions about the concept activity depend mainly on the techniques used to build models for activity recognition. We propose then a general paradigm to de ne a conceptual notion of human activity that is not subject to a particular application and which considers three levels of abstraction: at level 0 timed observations or discrete event, at level 1 primary activities as speci c timed observation sequences, and nally activity as sequences of primary activities. Besides, we present a general procedure to de ne activity models which combines Knowledge Engineering with Data Mining. The advantage of our approach is to facilitate the de nition of the principles of a general abstraction process from data. 3. A THEORETICAL FRAMEWORK FOR

MODELING HUMAN ACTIVITIES Our proposal is based on relating a Knowledge Engineering Methodology to a Timed Data Mining technique, i.e. the Timed Observations Modeling For Diagnosis (TOM4D) methodology [ 22, 14, 13 ] and the machine learning process called Timed Observations Mining For Learning (TOM4L) [ 4, 12 ]. Both TOM4D and TOM4L come from the mathematical Theory of Timed Observations [ 11 ] that provides a theoretical framework to facilitate the dynamic process modeling for monitoring, diagnosis and control. 3.1 Human Activities As Observation Classes In this framework, a process is an arbitrary set X(t) = fxi(t)gi=1:::r of time functions xi(t) de ned on R (i.e. sig

2 2 2 2 1 L =<X , Δ , C , M >

1 1 1 1 0

L =<X , Δ , C , M > Abstraction Level 0

L0=<X0, Δ0, C0, ∅> Θ(X2, Δ2)

Ω1 Θ(X1, Δ1) Ω0 nals provided by sensors). A timed observation is a couple ( i; tk) which corresponds to the assignation of a predicate (xi; i; tk) where i is constant and tk 2 R a time stamp. When making an abuse of language, such a predicate can always be interpreted as the predicate EQU ALS(xi; i; tk) (i.e. xi(tk) = i). A monitoring program (X; ) is a program that analyzes the set of time functions xi(t) associated to the set of variables X = fxigi=1:::r. The aim of a monitoring program is to write timed observations ( i; tk) in a database whenever a time function xi(t) 2 X(t) satis es some predicate (:; :; :). Generally speaking, such a predicate is satis ed when xi(t) matches against a behavioral model [ 10 ] that can be as simple as the switch of an interrupter or, requiring complex techniques, such as signal processing techniques for arti cial vision.

De nition 1. Let X be a set of variable names of a process X(t) = fxi(t)gi=1:::r and let = S x be such that x is a set of values assumable by thex2vXariable x 2 X via a program . An observation class Ci is a set of pairs (x; ) such that x 2 X ^ 2 x.

In other words, an observation class Ci associates variables x 2 X with constants 2 x. For simplicity reasons, an observation class is usually de ned as a singleton Ci = f(x; )g. This allows to formally de ne usual notions of events: A discrete event is a pair (x; ) with x 2 X; 2 x, denoting that the value is assumed by the variable x. A discrete event corresponds then to a singleton observation class Ci = f(x; )g.

A discrete event occurrence is a triplet (x; ; tk) with x 2 X; 2 x; tk 2 R denoting that the value is assumed by the variable x at the time tk. A discrete event occurrence is then a timed observation ( ; tk) of an singleton observation class Ci = f(x; )g.

The notion of observation class also contemplates di erent levels of abstractions; that is to say, a particular set C` = fC1`; :::; Cn`g; n 2 N of observation classes Ci` can be de ned for any level of abstraction ` (` 2 N). Thus, the following de nitions are introduced.

De nition 2. Let X` be a set of abstract variables belonging to an abstraction level ` and let ` = Sx`2X` `x` be such that `x` is a set of values assumable by the variable x`. An abstract observation class at the abstraction level ` is a singleton Ci` = f(x`; `)g, with x` 2 X` and ` 2 `x` .

De nition 3. A behavioral model M ` de ned at abstraction level ` is a set of n-ary timed relations between observation classes de ned at the abstraction level `.

The move from an abstraction level ` 1 to the level ` is made when associating a particular set of behavioral models (at level ` 1) to a given observation class Ci` (at level `). Considering this as a general principle, De nition 4 speci es the notion of abstraction level.

De nition 4. An abstraction level ` is a structure L` =< X`; `; C`; M ` 1 > where

X` is a set of variable names de ned at level `,

` is a set of values assumable by the variables, C` is the set of observation classes belonging to the level `, such that each observation class Ci` 2 C` is a singleton and,

M ` 1 is a behavioral model de ned at level `

Consequently, the variables X = fxigi=1:::r of a process X(t) = fxi(t)gi=1:::r are associated with the lowest level 0: L0 =< X0; 0; C0; M 1 > where X0 = X and, naturally, M 1 = ; since there is not observation classes in a previous level and so no behavioral models can be dened with the timed observation paradigm. At level 1, (∃ Lℓ+1) Θ(Xℓ+1, Δℓ+1)

Lℓ+1=<Xℓ+1, Δℓ+1, Cℓ+1, Mℓ> Knowledge

Base Expert L1 =< X1; 1; C1; M 0 > where the variables and the observation classes are abstract. Each class of C1 is associated with a behavioral model of M 0; that is, a set of nary relations included in M 0. Similarly, at level 2 where L2 =< X2; 2; C2; M 1 >, variables and observation classes are abstract and each observation class of C2 is associated with a sub-set of M 1.

The de nition of these abstraction levels allows to specify: the di erent types of discrete events (sensor data) as observation classes at level 0 (Ci0 2 C0), each primary activity as an observation class at level 1 (Ci1 2 C1) and; nally, an activity as an observation class at level 2 (Ci2 2 C2). The passage of a level ` 1 to a level `, where each class of C` is associated with a behavioral model of M ` 1, can be accomplished by a program (X`; `) which analyzes the ow of timed observations at level ` 1. In other words, (X`; `) assumes the matching of the ow of timed observations at level ` 1 against the models in M ` 1, and records the corresponding timed observation ( `; tk) in a database. Figure 4 illustrates these concepts in the context of the GerHome project where a monitoring program 0 registers a set 0 of sequences of discrete events, considered as timed observations at level 0, from sensors that perceive the process "the resident's behavior at home". For its part, (X`; `) (` = 1; 2) writes occurrences of observation classes de ned at level `, from a model M ` 1 that allows to recognize the behavior at level ` 1 and interpret it as more abstract activities at the higher level. 3.2 Activity Model Definition Process The abstraction process requires to establish the di erent levels; and therefore, to de ne behavioral models M `. To this aim, we propose a procedure of activity de nition based on the combination of learning from data using Data Mining techniques (TOM4L process), and the use of expert's knowledge through Knowledge Engineering (TOM4D methodology). Figure 6 shows the logic-precedence structure of the process of model construction where the relations are between passive entities (as knowledge base, process model and timed observations) and conceptually active entities (as TOM4D, TOM4L, the expert and the monitoring program ). Thus, an passive entity can be obtained trough an active entity which can require another passive entity. In the gure, a model can be built from a priori knowledge by means Knowledge

Base TOM4D

Process Model

TOM4L

Expert £(X,¢)

Timed Observation

Sequences of the methodology TOM4D or from timed observations in a database through TOM4L. Knowledge can come from experts' knowledge or can be new knowledge acquired from the built models validated by experts. In turn, the timed observations are obtained through a monitoring program which uses models to detect changes in the process and thus, it writes timed observations in a database. This structure of logic precedence allows to organize the available elements to carry out a procedure of model de nition.

Figure 5 illustrates the process of de nition of the inhabitant's activities for the GerHome project, where an monitoring agent 0 produces the timed observations in 0 (i.e. coming from sensors). The application of the TOM4L process (through the software ElpLab) to timed observations in ` (at level `) produces a behavioral model MT` OM4L representative of these observations. This model is analyzed through the TOM4D methodology and a source of knowledge (documents, data, experts, etc.) in order to de ne a behavioral model of interest M ` MT` OM4L and the abstract observation classes linked to this one (i.e. activities at the next level ` + 1). Thus, the abstraction level L`+1 =< X`+1; `+1; C`+1; M ` > can be speci ed in order that an agent detects in ` occurrences of M ` and registers occurrences of observation classes `+1. In a similar way, a new application of TOM4L on `+1 begins the cycle to de ne the activities of the next level which are later validated by experts.

The TOM4L process provides both a general matching program (X`; `) and a general algorithm to discover models M ` at any abstraction level. The next section illustrates the application of the TOM4L process to the GerHome project.

4. APPLICATION

Previous works on GerHome implement systems for monitoring elderly activities from video event and environment event [ 33, 34, 35 ]. The rst e orts to de ne activity models were carried out in a manual way from scenarios dened by experts. Nevertheless, the randomness of human behavior leads to that the manual de nition of activities is extremely complex. Consequently, we aim to de ne activities by means of using the TOM4L automatic techniques combined with available knowledge interpreted trough the TOM4D methodology.

Activity de nition is accomplished from data registered by sensors in the laboratory GerHome. This laboratory is an apartment made up living room, bathroom, kitchen and room, where di erent kinds of sensors record inhabitant behavior (Figure 7).

N E H C T I K

LIVING

ROOM

ROOM

BATHROOM

Sensors temperature, humidity, luminosity volumetric presence occupation (bed, chair,...) water consumption electricity consumption opening (doors, windows) opening (furnitures) image data concentrator GerHome's logs, as depicted in Figure 8, are timed data of the form "yymmdd-hhmmss.mss/Msg" where "yymmddhhmmss.mss" (like 080313-122225.825) is a time stamp tk and "Msg" (like USAGE.KITCHEN.MICRO_WAVE_OWEN.begin), is a constant associated with an observation class Ci0. The ElpLab software, which implements the TOM4L approach, uses a natural number i to identify the class. [...] 080313-122225.825/USAGE/KITCHEN.MICRO_WAVE_OWEN/begin 080313-122226.145/OPENCLOSE/KITCHEN.REFRIGERATOR/open 080313-122228.929/OPENCLOSE/KITCHEN.REFRIGERATOR/close [...] In particular, a spatial taxonomy is considered for the purpose of analyzing behavior in each area of home, so logs are classi ed according to the di erent spaces (Figure 7). In this section we describe how the activity de nition can be carried out by means of complementing knowledge about activities with data analysis. On the one hand, from a priori knowledge, the di erent abstraction levels of an activity can be speci ed; and thus, to analyze if the activity is representative of the available data. On the other hand, to analyze the available data to extract behavioral models and then, to de ne activities at di erent levels of abstraction. 4.1 From a priori Activity Definition to Data

Analysis Documents, set of data, information transmitted by experts and common sense allow to interpret sensor signals and to de ne what sequences of events determine an activity. Once this established, each activity can be validate by experts and collated with the resident's behavior registered in a database. Activities in the living room will be considered in order to illustrate part of the process of activity de nition in which the starting point is a priori knowledge. The living room of GerHome has ve sensors registering the resident's behavior: three detecting presence and other two detecting use of the phone and use of the TV. Each message of a timed data registered from sensors is interpreted through TOM4D as a variable that assumes a particular value; that is to say, as an observation class. For example, the messages PRESENCE.LIVING_ROOM.CHAIR.1.true and PRESENCE.LIVING_ROOM.CHAIR.1.false are interpreted 0 as a binary variable xL1 (PRESENCE.LIVING_ROOM.CHAIR.1) that takes values true or f alse. Thus, the observation classes C10027 = f(xL1; f alse)g and C10028 = f(xL1; true)g can be speci ed. Similarly, the other variables in the living room 0 0 are identi ed: xL2 (PRESENCE.LIVING_ROOM.CHAIR.2), xL3 0 (PRESENCE.LIVING_ROOM.ARMCHAIR), xL4 (USE.LIVING_ 0 ROOM.TEL) and xL5 (USE.LIVING_ROOM.TV); and so also, the corresponding observation classes: C10029 = f(x0L2; f alse)g, C10030 = f(x0L2; true)g, C10025 = f(x0L3; f alse)g, C10026 = f(x0L3; true)g, C10037 = f(x0L4; begin)g, C10038 = f(x0L4; end)g, C10039 = f(x0L5; begin)g and C10040 = f(x0L5; end)g.

Considering a priori knowledge on alternative activities in the living room and a certain notion on them, watch TV is proposed as a possible activity made up of sitting down and turning on the TV. Hence, an abstract class C1101 of level 1 can be speci ed to represent the activity watch TV and it can be associated with behavioral models of level 0, which are composed of at least the observation classes C10028; C10030, C10026 (linked to sit down) and C10039 (linked to turn on the TV). Therefore, although the models in principle are not known, some relation is supposed between the observation class 101 and the classes 1028; 1030; 1026; 1039 as Figure 9 illustrates.

From data and given a particular observation class, the TOM4L process allows to discovery the behavioral sequences that nish in the given class, and to nd the time constraints [0; 2 ] where 1 is the average times between two observation class occurrences. Then, taking in account the relations that would de ne watch TV (Figure 9), the study of the class 1039 (to turn on the TV) is carried out.

Figure 10 shows the behavioral model associated with the class C10039 where the discovered model consists only of turning on and turning o the TV. This indicates that only the 0 variable xL5 is involved in the behavior; that is to say, only the use of the TV as Figure 12(a) graphics (where true and Level 2

A similar result on the use of the armchair is obtained by studying the observation class C10026 where the found behavior consists only of sitting down and getting up of the armchair (Figure 11). Once again, there is only one variable (x0L3) involved in the discovered behavior as Figure 12(b) graphics.

These outcomes do not represent the intuitive idea about the behavior in a living room where if the resident actuates in the environment should exist some relation between the di erent variables (or sensors).

TOM4L de nes the BJ-measure [ 12 ] that allows to establish how strong is the relationship between the di erent observation classes. Figure 13 shows this measure between the observation classes of the living room, calculated from the available data, where columns and rows identify the mentioned classes. Note that the relations that exist are only those between the classes linked with the same variable. This explains the previously obtained models (Figures 10, 11) and allows to suppose that the available logs are not representative of a real-life watch TV activity.

The experts validated this deduction and thus, the robustness of the TOM4L approach was veri ed. Therefore, an a priori de nition of activity can be proposed by experts and collated with data in order to establish the adequacy of its de nition. 4.2 From Data Analysis to Activity Definition For the analysis of data, behavior executed in the kitchen is considered where there are 14 sensors and thus, 24 observation classes. The study is concerned with the use of the stove (PRESENCE.KITCHEN.STOVE.true, classID=1024). The TOM4L process provides a set of 50 n-ary relations that describes the occurrences of the class 1024 (Figure 14). The gure 15 shows one of these 50 n-ary relations. The proposal is then to use these relations in order to de ne activities in other level of abstraction.

The n-ary relations and their observation classes are analyzed and then grouped with di erent criteria de ned from the mentioned analysis, according to the TOM4D methodology. For example, the model m10 that describes the behavior OPENCLOSE.KITCHEN.CUPBOARD.SINK.open PRESENCE.KITCHEN.STOVE.true OPENCLOSE.KITCHEN.CUPBOARD.SINK.close OPENCLOSE.KITCHEN.REFRIGERATOR.close associated with using the stove is given in Figure 16, and an activity identi ed as A1, is associated with this model. An abstract observation class, let us say 111, representing this activity is speci ed and is linked with the aforesaid model. Table 2 shows di erent activities and their abstract classes speci ed de ning the level L1 =< X1; 1; C1; M 0 >. ElpLab allows to record the occurrences of the classes C1 of the level L1, and the same approach can be done to de ne L2 =< X2; 2; C2; M 1 >.

5. CONCLUSION

In this paper, a general theoretical framework to model and recognize resident activities was presented. Based on the areas of Knowledge Engineering and Timed Data Mining, this framework conceives human activities as entities at different levels of abstraction and generalizes thus, the notion of activity. This generalization allows that the de nitions of resident activity and the process of activity recognition are independent of any Data Mining technique or particular implementation. Besides, a general procedure to de ne the di erent abstraction levels and their behavioral models from data and experts' knowledge was described.

We applied our proposal to the GerHome's timed data coming from the sensors of a home prototype, in order to show that a priori Expert's knowledge can be collated with the timed data of a data base and, inversely, when a priori Expert's knowledge is not available, behavioral models can be found from timed data and then validated by Experts. We are now applying our approach to homes where activities are made by residents in di erent real-life context such as hospital or nursing room.

References

[1] AARP. Home and community preferences of the 45+ population . November 2010 .

[2] G. D. Abowd , A. F. Bobick , I. A. Essa , E. D. Mynatt , and W. A. Rogers . Discovering expert's knowledge from sequences of discrete event class occurrences . In ICEIS (2) , pages 253 { 260 , 2008 .

[3] F. K. Aldrich . Smart homes: past, present and future. Inside the smart home . Springer-Verlag, 2003 .

[4] N. Benayadi . Contribution a la decouverte de connaissances temporelles A~ a partir de donnees datees . Thesis , 2010 .

[5] B. Brumitt , B. Meyers , J. Krumm , A. Kern , and S. Shafer . Easyliving: Technologies for intelligent environments . In Proceedings of Second International Symposium on Handheld and Ubiquitous Computing , pages 12 { 29 , 2000 .

[6] D. Cook . Health monitoring and assistance to support aging in place . j-jucs , 12 ( 1 ): 15 { 29 , 2006 .

[7] D. Cook , J. Augusto , and V. Jakkula . Ambient intelligence: technologies, applications, and opportunities . Pervasive and Mobile Computing , 5 ( 4 ): 277 { 298 , 2009 .

[8] D. J. Cook and S. Das . Smart environments: technology, protocols and applications . John Wiley & Sons, Inc., 2005 .

[9] D. J. Cook , G. M. Youngblood , and G. Jain . Algorithms for smart spaces. The engineering handbook on smart technology for aging, disability and independence . A. Helal , M. Mokhtari and B. Abdulrazak . John Wiley & Sons, Inc., 2008 .

[10] L. Goc. SACHEM , a real-time intelligent diagnosis system based on the discrete event paradigm . Simulation , 80 ( 11 ): 591 { 617 , 2004 .

[11] M. L. Goc . Notion d'observation pour le diagnostic des processus dynamiques: Application a sachem et a la decouverte de connaissances temporelles. Habilitation a Diriger des Recherches . Unversite de Droit d' Economie et des Sciences d'Aix-Marseille , 2006 .

[12] M. L. Goc and N. Benayadi . Discovering expert's knowledge from sequences of discrete event class occurrences . In ICEIS (2) , pages 253 { 260 , 2008 .

[13] M. L. Goc and E. Masse . Towards a Multimodeling Approach of Dynamic Systems for Diagnosis . In Proceedings of the 2nd International Conference on Software and Data Technologies (ICSoft'07) , July 2007 .

[14] M. L. Goc , E. Masse, and C. Curt . Modeling Process From Timed Observations . In Proceedings of the 3rd International Conference on Software and Data Technologies (ICSoft'08) , July 2008 .

[15] J. B. H. Medjahed , D. Istrate and B. Dorizzi . Human activities of daily living recognition using fuzzy logic for elderly home monitoring . FUZZ-IEEE 2009. IEEE International Conference on Fuzzy Systems , pages 2001 { 2010 , 2009 .

[16] H. Hagras , F. Doctor , V. Callaghan , and A. Lopez . An incremental adaptive life long learning approach for type-2 fuzzy embedded agents in ambient intelligent environments . IEEE Transactions on Fuzzy Systems , 15 ( 1 ): 41 { 55 , 2007 .

[17] V. Jakkula , D. J. Cook , and A. Crandal . Enhancing anomaly detection for smart homes using temporal pattern discovery . Advanced intelligent systems . Springer, 2008 .

[18] E. Kim , S. Helal , and D. Cook . Human activity recognition and pattern ddiscovery . IEEE Pervasive Computing , 9 : 48 { 53 , 2010 .

[19] H. Muhammad . A computational framework for unsupervised analysis of everyday human activities . College of Computing Georgia Institute of Technology , 2008 .

[20] A. S. Ogale , A. Karapurkar , and Y. Aloimonos . View invariant modeling and recognition of human actions using grammars . Workshop on dynamical vision at ICCV05 . In In WDV, 2005 .

[21] M. Pantic , A. Pentland , A. Nijholt , and T. S. Huang . Human computing and machine understanding of human behavior: a survey . In Arti cal Intelligence for Human Computing , volume 4451 of Lecture Notes in Computer Science, pages 47 { 71 . Springer, 2007 .

[22] L. Pomponio and M. L. Goc . Timed observations modelling for diagnosis methodology: a case study . In Proceedings of the 5th International Conference on Software and Data Technologies (ICSoft 2010 ), July 2010 .

[23] P. Rashidi and D. Cook . Adapting to resident preferences in smart home . In AAAI Workshop on Advances in Preference Handling , pages 78 { 84 , 2008 .

[24] P. Rashidi and D. Cook . An adaptive sensor mining model for pervasive computing applications . In 2nd International Workshop on Knowledge Discovery from Sensor Data , 2008 .

[25] P. Rashidi and D. Cook . Keeping the resident in the loop: adapting the smart home to the user . Trans. Sys. Man Cyber. Part A , 39 ( 5 ): 949 { 959 , 2009 .

[26] P. Rashidi and D. J. Cook . Keeping the intelligent environment resident in the loop . In 4th International Conference on Intelligent Environments , pages 1 {9 , 2008 .

[27] P. Rashidi and D. J. Cook . An adaptive sensor mining framework for pervasive computing applications . Lecture Notes in Nomputer Ncince , 5840 : 154 { 174 , 2010 .

[28] P. Rashidi , S. Member , D. J. Cook , L. B. Holder , and M. Schmitter-Edgecombe . Discovering activities to recognize and track in a smart environment . IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING , 2010 .

[29] S. D. Santos and Y. Makdessi . Une approche de l' autonomie chez les adultes et les personnes a^gees. Etudes et resultats . DRESS, ( 718 ), February 2010 .

[30] R. Srikant and R. Agrawal . Mining sequential patterns: generalizations and performance improvements . In 5th International Conference on Extending Database Technology. Advances in Database Technology EDBT 96. , pages 3 { 17 , 1996 .

[31] S. H. You , H. J. Park , T. S. Kim , J. W. Park , U. Burn, J. A. Seol , and W. D. Cho . Developing intelligent smart home by utilizing community computing . In UCS'07: Proceedings of the 4th International Conference on Ubiquitous Computing Systems , pages 59 { 71 , Berlin, Heidelberg, 2007 . Springer-Verlag.

[32] G. M. Youngblood and D. J. Cook . Data mining for hierarchical model creation . IEEE Transactions on Systems - Man, and Cybernetics . Part C: Applications and Reviews, 37 ( 4 ): 561 { 572 , 2007 .

[33] N. Zouba , B. Boulay , F. Bremond , and M. Thonnat . Cognitive vision. chapter Monitoring Activities of Daily Living (ADLs) of Elderly Based on 3D Key Human Postures , pages 37 { 50 . Springer-Verlag, 2008 .

[34] N. Zouba , F. Bremond , and M. Thonnat . Multisensor fusion for monitoring elderly activities at home . In Proceedings of the 2009 Sixth IEEE International Conference on Advanced Video and Signal Based Surveillance, AVSS '09 , pages 98 { 103 . IEEE Computer Society, 2009 .

[35] N. Zouba , F. Bremond , M. Thonnat , A. Anfosso , E. Pascual, P. Mallea , V. Mailland , and O. Guerin . A computer system to monitor older adults at home: Preliminary results . Gerontology. International journal on the fundamental aspects of technology to serve the ageing society , 8 ( 3 ), Summer 2009 .