Generic Distributed Sensing in Support of Context Awareness in ...

4 downloads 4493 Views 443KB Size Report
Cite this paper as: Xiao B., Kanter T., Rahmani R. (2014) Generic Distributed Sensing in Support of Context Awareness in Ambient Assisted Living. In: Park J.
Generic Distributed Sensing in Support of Context Awareness in Ambient Assisted Living Bin Xiao*, Theo Kanter, and Rahim Rahmani Department of Computer and Systems Sciences, Stockholm University Forum 100, Isafjordsgatan 39, SE-164 40 Kista, Sweden {xbin,kanter,rahim}@dsv.su.se

Abstract. Researches in ambient assisted living have so far faced three important challenges: (1) Lack of a comprehensive approach to capture user needs that are generic; i.e., not limited to specific events, but as generic related to the user. (2) Lack of a highly flexible and scalable platform for the distributed sharing and processing of context between nodes in IoT networks. (3) Increased amount of communication and devices with sensors participating in the acquisition, processing and sharing of context further challenges both computation capability and storage capacity of the system. In this paper, we address these limitations and present novel support, applied in a system for remote assistance of elderly. The support comprehensively retrieves user needs from generic context, via a scalable overlay providing increment of processing capability and storage. Further, the support self-organizes entities into generic context from distributed sensing, using the Dependent Context Pattern (DCP) based on the Context Virtualizing Platform (CVP). Keywords: Generic Context, Context Dependency, User Needs, Virtualization Platform.

1

Introduction

With the development of global society aging, more and more elderly suffering from slight dementia needs to be cared or supported in daily life. Researchers utilize sensor technology and multi-media to make they live independently, called Ambient Assisted Living (AAL) [1]. In recent research, the AAL service shouldn’t just aim at long-term monitoring. Instead, researchers need to focus more on capturing what the users interested in, and provide a dynamic system adaption [2]. Utilizing heterogeneous sensors to detect the context related to users and making an intelligent responding to the users (such as reasoning) are the primary paradigm for contemporary research. However, even sensors have advantages like: objective, easyto-control, and stable, but conventional physical sensors are not powerful enough to fully capture the flexible and comprehensive context of users [3]. In the conventional view of ambient intelligence, researchers emphasize too much on the physical side of *

Corresponding author.

James J. (Jong Hyuk) Park et al. (eds.), Multimedia and Ubiquitous Engineering, Lecture Notes in Electrical Engineering 308, DOI: 10.1007/978-3-642-54900-7_14, © Springer-Verlag Berlin Heidelberg 2014

99

100

B. Xiao, T. Kanter, and R. Rahmani

environment for user context, such as light, temperature, heat, etc.. But the user context in AAL should not be just limited by this, because users live under a quick changing human surroundings with all kinds of events happening. For example, early this morning, the subway is stopped since a terribly heavy snow. But the elderly planned to go outside for shopping. With a slight dementia, she may not know how to deal with that sudden event. Of course, we can expect caregivers to give her a call as reminder, but it is not security-guaranteed. Because the situation is changing quickly with quite many potential problems and some of them not even be recognized by the caregivers. For such case, the “context” we need to acquire is not just the physical status inside the room. Rather than that, we need a broad channel to capture context dispersed in the human surroundings. On the other hand, with the quickly increasing nodes amount in the system, system computation pressure increased quickly. For this issue, most researchers make compromise between a centralized infrastructure and self-support infrastructure [4]. The centralized infrastructure means system owns a central point to schedule nodes and processes the context aggregately. The self-supported infrastructure indicates that some context should be locally processed by the nodes but the processed results are shared with other peers. In that case, local nodes must be capable enough in some extent for context processing, which introduces a problem of “heavy client”. However, a fully distributed and virtualized infrastructure break the impasse of centralized or self-supported, which allows the nodes processing the context in a fully distributed platform, without any central point, without local heavy context processing [5]. The paper will be formed like this. Chapter 2 propose three challenges for AAL research. In Chapter 3, section 3.1 will discuss the Generic Context (GC) sensing and the Dependent Context Pattern (DCP) to solve the challenge one, while section 3.2 will propose the Context Virtual Platform (CVP) to solve challenge two and three. Chapter 4 will carry out model checking. Chapter 5 will draw a conclusion.

2

Research Challenges for AAL

It is an important research problem in AAL to capture user needs and make system adapt to that, in the view of most researchers [6]. For user needs mining, we need a comprehensive and full approach to capture user context. However, most available indoor AAL systems only emphasize on physical factors in the house (e.g. user gesture, light, temperature, etc.) [7] [8] [9]. Partial attention on using physical sensors to acquire context leads some blind area of human surroundings and makes the context limited and inflexible. Except the conventional context for physical environment, some implicit context hiding in human surrounding must be discovered, such as some accidents or disasters happening around the user, weather changing, traffic situation, and even the supermarket discount. With contribution from those contexts the system can capture user needs more humanly and precisely. For instance, with knowing of the coming thunder storm, system can remind the elderly with dementia to wear more clothes or take umbrella before leaving home [10]. By aggregating context from physical environment and human surrounding, we form a Generic Context (GC), which

Generic Distributed Sensing in Support of Context Awareness

101

makes the system deduce user needs more precisely and specifically, even acquiring the user social needs. It will be discussed in details later. However, no matter the physical environment context or the Generic Context (GC, aggregation of physical environment and human surrounding) is highly dispersed and existing with different nature, which is difficult to acquire by only using conventional physical sensors. GC describes the quick changing surroundings using the Dependent Context Pattern (DCP), to which conventional sensors are not flexible enough to track and detect. Moreover, since the context existing discretely, we have to associate each context fragment for context awareness, where the DCP works. Actually, what we should aware of are patterns of the context changing and their relations, which reflecting the real world. As compensation to the physical environment context, human are involved in the system to contribute for human surrounding context. Human is more flexible and capable than physical sensors to precisely capture some context by crowdsourcing. Fig.1 (left) shows the paradigm of retrieving user needs from the GC. Human participation keeps acquiring and sharing the human surrounding context. The context contributed from human is irreplaceable which compensate to the physical environment. By associating things and events retrieved from the GC, user needs can be comprehensively exposed. As a short conclusion, we introduce the first challenge needs to be solved in our research. Challenge one: Need a comprehensive approach to fully acquire the dispersed and flexible context. The context processing pressure increase quickly, because more and more context needs to be acquired, and more and more nodes are joining into the system. On the other hand, the context log grows quickly with time, which is very important to be kept for ambient intelligence computation. There are many platforms available for context collection, processing, and storage. But none of them is highly scalable and extensible [11]. Some of them are half-distributed, which introduce a central point for context processing and nodes scheduling. Those non-distributed and semi-distributed architecture makes system difficult to be extend and scaled. There exist two disadvantages for those systems: (1) Each fault in the central point could be deadly for the system. In quite many cases, a small error in the central point leads a system crash. (2)With the increasing of context waiting for processing, the system is facing with an increasing security risk and computation pressure. To avoid those disadvantages, we need to keep away from the centralized structure. As a short conclusion, challenge two is introduced as: Challenge two: need a scalable and extensible platform for retrieving, sharing and processing disperse context. Some available systems introduce specialized hardware or specialized machine in the nodes for local context acquiring, processing and storing, called “self-supported approach” [4]. They keep part of computation task locally and store some context locally, so as to alleviate the two problems (mentioned above) for the centralized topology. However, that approach puts requirement on context processing capability for the introduced hardware and devices, which may increase the system. As a short conclusion, challenge three is introduced as:

102

B. Xiao, T. Kanter, and R. Rahmani

Challenge three: increasing nodes and context log cause heavy context storage and processing pressures on both central point and other distributed nodes. The Context Virtual Platform (CVP) will be introduced to deal with the challenge two and challenge three, in next chapter. CVP is a fully distributed virtual platform based on the Mediasense, without any central point [11]. With more and more nodes joining into, CVP increase the system computation capability by integrating the resource of each node together, as a virtual platform. Thus the challenge two and three are solved. As the peer-to-peer bidirectional communication basis of CVP, Mediasense is a fully distributed platform for internet of things, which uses Distributed Context eXchange Protocol (DCXP) to share the data between each node seamlessly [11]. Different from other systems, it is lightweight, scalable, and extendable, without central point.

3

Paradigm, Pattern and Platform for Generic Context

3.1

Generic Sensing Approach and Dependent Context Pattern

Human surrounding context includes user profile, surrounding dynamics, and tasks [12]. User profile relates to social relations, user disease history, etc.. Surrounding dynamics means events happening and status related to the user (e.g. weather forecast, disaster forecast, etc.) [12]. Task means activities and goals users should engage into [12]. For the assisted living system, it is important to hold the context of human surrounding. User profile and user tasks play vital roles to decide how “smart” the system can react to user needs, such as health status, habit, and even social relations. With the context from both physical environment and human surroundings, we form the Generic Context (GC) to percept the user needs, as described in Fig.1 (right) [12]. In order to capture human surrounding context, human will participate into the sensing, as shown in Fig.1 (left). Human is more flexible to percept the uncertain and changing context. Also human can diagnose context value and filter some unwanted context [13]. This hybrid sensing approach forms a “human-in-the-loop sensing” [13], by which the system can process context contributed from the human participants by crowdsourcing, then give an attentive and repeatable reminder to the elderly user based on GC. In such way, the first challenge is solved.

Fig. 1. Generic Context Sensing (left) and Subdivision of Generic Context (right)

Generic Distributed Sensing in Support of Context Awareness

103

Fig. 2. Ontology based model for Generic Context--Dependent Context Pattern

In this article, we introduce the Dependent Context Pattern to model user context, with three dimensions (i.e. time, dependence, features), by which we define the context entity with a unique ID, composed by group of values on each dimension. Each entity is under certain kind of relations with other entities. Entities with similar features form a cluster, based on which a relation net is established, allowing indexing, query and processing. Fig.2 shows the Dependent Context Pattern (DCP). The DCP will be used to serve for Generic Context sensing by its five-layer model, where each layer provides service for the upper layer, shown in Fig.3 (right). The observation layer observes physical features of the environment by sensors, which act as system boundary interacting with the world by sensors, such as using Kinect to observe the shape and color of artifacts. Based upon the observation layer, metadata will be delivered to the perception layer, in which features are clustered to form a space of context entity. In the entity layer, system will index and make a matching between the existing entities with their features and the captured features. And create a new entity for the features, who cannot match any existing entity. The semantic layer builds relations among entities by machine learning. With those relations, the context storage is formed. The context layer will output context service using the semantic layer as its input. With the support from DCP, we can comprehensively and fully capture the context for user needs by following a GC sensing way. 3.2

The Virtual Platform for General Context

In this section, we introduce a Context Virtual Platform (CVP) to give an infrastructure support for the layer model DCP, in the view of infrastructure. We try to avoid introducing any central point in the platform to deal with the challenge two and three, by using a completely distributed structure, based on the DCP. In CVP, each node is

104

B. Xiao, T. Kanter, and R. Rahmani

contributed to process the context on the platform logically, rather than keeping major computation task locally. However, we still hope to carry out the context computation centrally in logic, which benefits for the context flowing and sharing between nodes. Thus CVP provides a virtual platform, which processes the context centrally but distributed in structure. The CVP is a virtualized infrastructure based on Mediasense--an IoT platform which allows bi-directional peer-to-peer communication between nodes [11]. CVP follows the idea of “shared-memory and symmetric multi-processing model” using hypervisors [14]. Based on the Mediasense which provide seamless communication among nodes [11], CVP provides a hypervisor mechanism where each node is capable to join the centrally context processing and cooperated to form a virtual central processor as shown in Fig. 3 (left). In each node, a Raspberry Pi owns its private memory and processor. Under the coordinating of CVP, each node will share part of their private memory and processor with the virtual platform to form a global context storage and a logic-existing central processor. In such way, with more nodes joining into the platform, the virtual storage and virtual processing capability will steady grow, regardless it is in logarithmic increasing or in linear increasing. So CVP makes the system easy to be scaled and extended, with increasing system capability to deal with the increasing system pressure from each node and the central point. Thus, we can solve the challenge two and three by this CVP in support of DCP.

Fig. 3. Context Virtual Platform (left) Supports Layered Dependent Context Pattern (right)

4

Evaluation

We use model checking to verify logic correctness of the proposed models, through the transition systems description for each layer in the layered DCP model [15]. The DCP layer model hybrids process logic in CVP to support the DCP. The transition

Generic Distributed Sensing in Support of Context Awareness

105

system is defined as M = ( S , →, L ) [16]. S is a group of states, → is the relations between states, and L is a labeling function [16]. Then we declare the system properties ϕ for each layer and check whether M |= ϕ . The labeling algorithm is used in this model checking, where the defined transition models and properties are input then states satisfying with ϕ are output [17]. Computational Tree Logic (CTL) is used to describe properties of the branching system. CTL provides a semantic to describe both the time features and system path. “A” and “E” are path variables, while “X”, “F”, “G”, and “U” are time variables [16]. “A” means for all the paths, “E” means existing a path. “X” means in the next state, “F” means in the future, “G” means globally in the future, “U” means until. System properties are expressed, using the combination of path variables and time variable. Table 1. Properties and Proposition of model checking Properties and Proposition Formulas

Layer Observe Percept

Object Semantic Context

{D:distiguished, F:feature-seperated, A:active, P: positive} φ 1: AG((D→EF F) & EU(P,A)); φ2: AG(!(P → EF A )& !(F → EF D)). {Cla:calssification, Clu:cluster-processed, Spe:special-processing, A: active, P: positive} φ 3: AG(P→EX A); φ 4: AG (!(A →EF P) & !(Spe & A →EF (Clu|Cla))). {Exi: existed object( otherwise need a new object); TF: tagged feature, A: active, P: positive;}φ 5: AG(Exi & P→EF (TF&A)); φ 6: AG((TF&A) → ∅). Ass: associated, In: index marked, A: active, P: postive; φ 7 : AG((Ass | P→EF In & A) &(P →EF A)); C: context, Ser:serving, K: knowledge, A: active, P: postive φ 8: AG (!∅ → EF (S & A)); φ 9: AG !(S →EF (K | C)).

The properties and propositions for each layer are defined in Table 1. The properties (φ) are defined based on system robustness, liveness, and accuracy. Propositions (like Exi, D, F etc.) indicate the amount of S and features for each S in each layer. We use the NuSMV as our model checking tool. NuSMV is a symbolic tool, based on SMV [18]. We describe each layer in the layered DCP using SMV model language [18], according to the propositions defined in Table 1. Then we input all the properties in Table 1, to pick out Sc where M |= ϕ , until S c ⇔ S . So we ensure the model quality and logic correctness by controlling the robustness, liveness, and accuracy.

5

Conclusion

The contribution of this paper starts at chapter 2, by defining three research challenges in AAL, which has no enough attention in current research. Aiming at the first challenge, we proposed DCP with GC sensing in Section 3.1. By integrating DCP with GC sensing, we can fully capture the user needs in a comprehensive way, where many researchers did not pay enough attention to in previous research. In Section 3.2,

106

B. Xiao, T. Kanter, and R. Rahmani

we aim at challenge two and three to introduce the CVP, which supports DCP in infrastructure and form a highly scalable and flexible system for AAL by peer-to-peer communication. It deals with the challenge two and three, where very few researchers have formally provide specific or practical solution. By the prototype, we find our models can well fit for the AAL scenario case. Chapter 4 verifies the logic correctness for DCP and CVP, using Transition System and Computation Tree Logic. By the GC and DCP, we successfully expand the context acquiring channel for previous research. CVP deals with the embarrassing situation of compromise between centralized structure and self-support structure, by providing a flexible and scalable virtual platform with increasing processing capability and storage capacity, where few really practical solution are provided in previous research. Our current work focuses on defining the paradigm of DCP, CVP model, and Generic Context Sensing methodology. The future work is to create an implemented CVP prototype and DCP algorithm. Acknowledgement. Research funding from the European research project–Smart Assisted Living Involving Informal Care Givers (SALIG++) is deeply appreciated.

References 1. Chiriac, S., Saurer, B.R.: An Ambient Assisted Living Monitoring System for Activity Recognition - Results from the First Evaluation Stages. In: Fifth German AAL Congress, Berlin, Germany (2012) 2. Cook, D.J., Augusto, J.C., Jakkula, V.R.: Ambient intelligence: technologies, applications, and opportunities. Pervasive and Mobile Computing 5, 277–298 (2009) 3. Boulos, M.N.K., et al.: Crowdsourcing, citizen sensing and sensor web technologies for public and environmental health surveillance and crisis management: trends, OGC standards and application examples. International Journal of Health Geographics 10(1), 67 (2011) 4. Guo, B., et al.: From the internet of things to embedded intelligence. World Wide Web Journal, 1–22 (September 2012) 5. Fitton, D., Sundramoorthy, V., Kortuem, G., Brown, J., Efstratiou, C., Finney, J.: Exploring the Design of Pay-Per-Use Objects in the Construction Domain. In: Roggen, D., Lombriser, C., Tröster, G., Kortuem, G., Havinga, P. (eds.) EuroSSC 2008. LNCS, vol. 5279, pp. 192–205. Springer, Heidelberg (2008) 6. Emiliani, P.L., Stephanidis, C.: Universal access to ambient intelligence environments: Opportunities and challenges for people with disabilities. IBM Systems Journal 44(3), 605–619 (2005) 7. Sit, G.F., et al.: Application-Oriented Fusion and Aggregation of Sensor Data. In: Ambient Assisted Living, pp. 3–13. Springer, Heidelberg (2012) 8. Storf, H., Kleinberger, T., Becker, M., Schmitt, M., Bomarius, F., Prueckner, S.: An eventdriven approach to activity recognition in ambient assisted living. In: Tscheligi, M., de Ruyter, B., Markopoulus, P., Wichert, R., Mirlacher, T., Meschterjakov, A., Reitberger, W. (eds.) AmI 2009. LNCS, vol. 5859, pp. 123–132. Springer, Heidelberg (2009) 9. Becker, M., et al.: Approaching ambient intelligent home care systems. In: Pervasive Health Conference and Workshops. IEEE (2006)

Generic Distributed Sensing in Support of Context Awareness

107

10. Metso, A., et al.: Ubiquitous Services for Senior Citizens-Living Process Detection and Middleware. In: Eighth IEEE/ACIS International Conference on Computer and Information Science, ICIS 2009. IEEE (2009) 11. Kanter, T., et al.: Mediasense–an internet of things platform for scalable and decentralized context sharing and control (2012) 12. Schmidt, A., Beigl, M., Gellersen, H.-W.: There is more to context than location. Computers & Graphics 23(6), 893–901 (1999) 13. Sheth, A.: Citizen Sensing, Social Signals, and Enriching Human Experience. IEEE Internet Computing 13(14), 80–85 (2009) 14. Sobania, J.-A.: Memory Management in a Many-Core Distributed Hypervisor. Ph. D. Retreat of the HPI Research School an Service-oriented Systems Engineering, 165 15. Wang, J., He, X., Deng, Y.: Introducing software architecture specification and analysis in SAM through an example. Information and Software Technology 41(7), 451–467 (1999) 16. Reeves, S., Clarke, M.: Logic for computer science. Addison-Wesley, Wokingham (1990) 17. Baier, C., Katoen, J.-P.: Principles of model checking, vol. 26202649. MIT Press, Cambridge (2008) 18. Cimatti, A., Clarke, E., Giunchiglia, E., Giunchiglia, F., Pistore, M., Roveri, M., Sebastiani, R., Tacchella, A.: NuSMV 2: An openSource tool for symbolic model checking. In: Brinksma, E., Larsen, K.G., et al. (eds.) CAV 2002. LNCS, vol. 2404, pp. 359–364. Springer, Heidelberg (2002)

Suggest Documents