1

Internet of Things (IoT), a Roadmap for Smart Environments

Khachik Babaians*

ARICELL Co., Unit no. 60, 12th floor, building No. 21.4, Arghutyan str. 2nd alley, Griboyedov

St., Komitas Ave., Zip code: 0051 Yerevan/Armenia

1 IntroductionEvery day we create 2.5 quintillion bytes of data; so much that 90% of the data

in the world today has been created in the last 2 years alone. This data comes fromsensors used to gather climate information, from posts to social media sites, digitalpictures and videos, purchase transaction records, or cell phone GPS signals, toname only a few. This data is Big Data. Analyzing large data sets alreadyunderpins new waves of productivity growth, innovation, and consumer surplus.Big data is more than simply a matter of size; it is an opportunity to find insights innew and emerging types of data and content, to make businesses more agile, and toanswer questions that were previously considered beyond our reach. Until now,there was no practical way to harvest this opportunity. But today we are witnessingan exponential growth in the volume and detail of data captured by enterprises, therise of multimedia, social media and Online Social Networks (OSN), and theInternet of Things (IoT).

Technology becomes more and more part of our daily life. New technologieshave finally reached a stage of development in which they can significantlyimprove our lives. For example, our cities are fast transforming into artificialecosystems of interconnected, interdependent intelligent digital “organisms”. Theyare transforming into smart cities, as they benefit more and more from intelligentapplications designed to drive a sustainable economic development and anincubator of innovation and transformation that merges the virtual world of MobileServices, IoT and OSN with the physical infrastructures of Smart Building, SmartUtilities (i.e., electricity, heating, water, waste, transportation, and unifiedcommunication and collaboration infrastructure).

* Corresponding author. Tel.: (+374)-41-777519

E-mail addresses: khachikbabaians@gmail.com; babaians@aricell.net (Khachik Babaians)

2

As part of the Future Internet, IoT aims to integrate, collect information from-,and offer services to a very diverse spectrum of physical things used in differentdomains. “Things” are everyday objects for which IoT offers a virtual presence onthe Internet, allocates a specific identity and virtual address, and adds capabilitiesto self-organize and communicate with other things without human intervention.To ensure a high quality of services, additional capabilities can be included such ascontext awareness, autonomy, and reactivity. Things are very diverse. Very simplethings, like books, can have Radio Frequency identification-RFID tags that helptracking them without human intervention. For example, in an electroniccommerce system, a RFID sensor network can detect when a thing left thewarehouse and can trigger specific actions like inventory update or customerrewarding for buying a high end product [1]. In this simple case, RFIDs enable theautomatic identification or things, the capture of their context (for example thelocation) and the execution of corresponding actions if necessary. Sensors andactuators are used to transform real things into virtual objects [2] with digitalidentities. In this way, things may communicate, interfere and collaborate witheach other over the Internet [3]. Adding part of application logic to thingstransforms them into smart objects, which have additional capabilities to sense, logand understand the events occurring in the physical environment, autonomouslyreact to context changes, and intercommunicate with other things and people. Atool endowed with such capabilities could register when and how the workers usedit and produce a financial cost figure. Similarly, smart objects used in the e-healthdomain could continuously monitor the status of a patient and adapt the therapyaccording to the needs. Smart objects can also be general purpose portable deviceslike smart phones and tablets, that have processing and storage capabilities, and areendowed with different types of sensors for time, position, temperature, etc. Bothspecialized and general purpose smart objects have the capability to interact withpeople.The main function of the IoT infrastructure is to support communication amongthings (and other entities such as people, applications, etc.). This function must beflexible and adapted to the large variety of things, from simple sensors tosophisticated smart objects. More specific, things need a communicationinfrastructure that is low-data-rate, low-power, and low-complexity. Actualsolutions are based on short-range radio frequency (RF) transmissions in ad-hoc

3

wireless personal area networks (WPANs). A main concern of the IoTinfrastructure developers is supporting heterogeneous things by adoptingappropriate standards for the physical and media access control (MAC) layers, andfor communication protocols. The protocol and compatible interconnection for thesimple wireless connectivity with relaxed throughput (2–250 kb/s), low range (upto 100 m), moderate latency (10–50 ms) requirements and low cost, adapted todevices previously not connected to the Internet were defined in IEEE 802.15.4.Other similar efforts refer to industrial and vehicular applications. The IoTarchitecture supports physical things’ integration in Internet and the complexinteraction flow of services triggered by events occurrence. The main conceptsinvolved are the object-identification, sensing and connecting capabilities as thebasis for the development of independent cooperative services and applicationsthat address several key features for IoT architecture: Service Orientation, Webbase, distributed processing, easy integration via native XML and SOAPmessaging, component-base, open access, N-tiered architecture, support forvertical and horizontal scalability [4]. These features allow “physical objectsbecome active participants in business processes” [5]. As a consequence, the WebServices must be available to interact with the corresponding virtual objects overthe Internet, query and change their state and any information associated withthem. The new key features for the IoT architecture include persistent messagingfor the highest availability, complete security and reliability for total control andcompliance, platform independence and interoperability (more specific formiddleware).Sensors in IoT can run anywhere and on any objects. They are used to collect datasuch as biomedical information, environment temperature, humidity, ambient noiselevel. This is also a research issue close to Big Data science. The data provided bysuch sensors can be used by customized context-aware applications and services,capable to adapting their behavior to their running environment. However, sensordata exhibits high complexity (e.g., because of huge volumes and inter-dependencyrelationships between sources), dynamism (e.g., updates performed in real-timeand data that can critical age until it becomes useless), accuracy, precision andtimeliness. An IoT system should not concern itself with the individual pieces ofsensor data: rather, the information should be interpreted into a higher, domain-relevant concept. For example, sensors might monitor temperature, humidity, whilethe information needed by a watering actuator might be that the environment is

4

dry. This higher-level concept is called a situation, which is an abstract state ofaffairs interesting to applications [6].

2 Context-Aware Infrastructures for the Internet of Things

A context-aware system is generally characterized by several functions. Itgenerally gathers context information available from user interface, pre-specifieddata or sensors and adds it to a repository (Context Acquisition and Sensing).Furthermore, the system converts the gathered raw context information into ameaningful context which can be used (Context Filtering and Modeling). Finally,the system uses the context to react and make the appropriate context available tothe user (Context Reasoning, Storage and Retrieval).

Being context-aware allows software not only to be able to deal with changesin the environment the software operates in, but also being able to improve theresponse to the use of the software. That means context-awareness techniques aimat supporting both functional and non-functional software requirements. Authors of[7] identified three important context-awareness behaviors:1. The representation of available information and services to an end user.2. The automatic execution of a service.3. The tagging and storing of context information for later retrieval.Sensors in IoT can run anywhere and on any objects. They are used to collect datasuch as biomedical information, environment temperature, humidity, ambient noiselevel. This is also a research issue close to Big Data science. The data provided bysuch sensors can be used by customized context-aware applications and services,capable to adapting their behavior to their running environment. However, sensordata exhibits high complexity (e.g., because of huge volumes and inter-dependencyrelationships between sources), dynamism (e.g., updates performed in real-timeand data that can critical age until it becomes useless), accuracy, precision andtimeliness.An IoT system should not concern itself with the individual pieces of sensor data:rather, the information should be interpreted into a higher, domain-relevantconcept. For example, sensors might monitor temperature, humidity, while theinformation needed by a watering actuator might be that the environment is dry.This higher-level concept is called a situation, which is an abstract state of affairsinteresting to applications [6].

5

The research topics on situation identification for IoT involve several issues [8].First, representation deals with how to define logic primitives used to construct asituation’s logical specification. In representation, logical primitives should capturefeatures in complex sensor data (e.g., acceleration data), domain knowledge (e.g., aspatial map or social network), and different relationships between situations. Also,an IoT system is assumed to be highly dynamic. New sensors can be introduced,that introduce new types of context. Therefore, the logical primitives should beflexibly extensive, such as new primitives to not cause modifications or produceambiguous meanings to existing ones. Specification deals with defining the logicbehind a particular situation. This can be acquired by experts or learned fromtraining data. It typically relies on a situation model with priory expert knowledge,on which reasoning is applied based on the input sensor data. For example, in logicprogramming [9] the key underlying assumption is that knowledge about situationscan be modularized or digitized.Other logic theories, such as situation calculus [10], have also been used to infersituations in IoT systems. Kalyan et al. [11] introduce a multi-level situationtheory, where an intermediate level micro situation is introduced between infonsand situations. An infon embodies a discrete unit of information for a single entity(e.g., customer or a product), while a situation makes certain infons factual andthus support facts.

Fig. 1 An example of situation inferring using Dempster-Shafer theory (from [12])

6

Other solutions are based on the Dempster-Shafer theory (DST) [13], amathematical theory of evidence, which propagates uncertainty values andconsequently provides an indication of the certainty of inferences. The process ofusing DST is described as follows. First, developers apply expert knowledge toconstruct an evidential network that describes how sensors lead to activities. Theleft-hand side of Fig. 1 describes that the sensors on the cup and fridge are

connected to context information (e.g.,‘cup used’). Such context information canbe further inferred or composed to higher-level context. The composition of

context information points to an activity (e.g.,‘get drink’) at the top. Developerscan use such an approach to determine the evidence space and degree of belief inevidence. For example, in the figure the values on the arrows represent the belief inparticular sensor (also called the uncertainty of sensor observations). Generally, inreasoning situations are inferred from a large amount of imperfect sensor data. Inreasoning, one of the main processes is called situation identification-deriving asituation by interpreting or fusing several pieces of context in some way.Specifying and identifying situations can have a large variability depending onfactors such as time, location, individual users, and working environments. Thismakes Specification-based approaches relying on models of a priori knowledgeimpractical to use. Machine learning techniques have been widely applied tolearning complex associations between situations and sensor data. However, theperformance of reasoning is usually undermined by the complexity of theunderlying sensor data.Bayesian networks and Hidden Markov Models (HMMs) have been applied inmany context-aware systems. In HMMs statistical models a system being modeledis assumed to be a Markov chain that is a sequence of events [41]. A HMM is com-posed of a finite set of hidden states and observations that are generated fromstates. For example, a HMM where each state represents a single activity (e.g.,

‘prepare dinner’,‘go to bed’,‘take shower’, and‘leave house’) is presented in

[14]. They represent observations in three types of characterized sensor data thatare generated in each activity, which are raw sensor data, the change of sensordata, the last observed sensor data, and the combination of them. The HMM istrained to obtain the three probability parameters, where the prior probability of anactivity represents the likelihood of the user starting from this activity; the statetransition probabilities represent the likelihood of the user changing from one

7

activity to another; and the observation emission probabilities represent thelikelihood of the occurrence of a sensor observation when the user is conducting acertain activity.Finally, Support Vector Machines (SVM) [15] is a relatively new method forclassifying both linear and nonlinear data. An SVM uses a nonlinear mapping totransform the original training data into a higher dimension. Within this newdimension, it searches for the linear optimal separating hyper-plane that separatesthe training data of one class from another. With an appropriate nonlinear mappingto a sufficiently high dimension, data from two classes can always be separated.SVMs are good at handling large feature spaces since they employ over fittingprotection, which does not necessarily depend on the number of features. Kanda etal. [16]

3 Functional Requirements

The ‘things’ configuration process detects, identifies, and configures sensor

hardware and cloud-based IoT platforms in such a way that software platforms canretrieve data from sensors when required. In this section, we identify theimportance, major challenges, and factors that need to be considered during aconfiguration process. The process of sensor configuration in IoT is important fortwo main reasons. Firstly, it establishes the connectivity between sensor hardwareand software systems which makes it possible to retrieve data from the deployedsensor. Secondly, it allows us to optimize the sensing and data communication byconsidering several factors as discussed below. Let us discuss the followingresearch problem: Why is sensor configuration challenging in the IoTenvironment? The major factors that make sensor configuration challenging are(1)the number of sensors, (2)heterogeneity, (3)scheduling, sampling rate,communication frequency, (4)data acquisition, (5) dynamicity, and (6) context[17].

8

Table 1 Heterogenecity in term of Wireless Communication Technology

Fig.2 Heterogeneity in item of sensing/measurement capabilities of sensor nodes

1. Number of Sensors: When the number of sensors that need to be configuredlimited, we can use manual or semi-autonomous techniques. However, when thenumbers grow rapidly towards millions and billions, as illustrated in Fig. 1b, suchmethods become extremely inefficient, expensive, labour-intensive, and in mostsituations impossible. Therefore, large numbers have made sensor configurationchallenging. An ideal sensor configuration approach should be able to configuresensors autonomously as well as within a very short time period.

2. Heterogeneity: This factor can be interpreted in different perspectives. (1)Heterogeneity in terms of the communication technologies used by the sensors, aspresented in Table 1. (2) Heterogeneity in terms of measurement capabilities, aspresented in Fig. 2 (e.g. temperature, humidity, motion, pressure). (3) The types ofdata (e.g. numerical (small in size), audio, video (large in size)) generated by thesensors are also heterogeneous. (4) The communication sequences and securitymechanisms used by different sensors are also heterogeneous (e.g. exactmessages/commands and the sequence that needs to be followed to successfully

9

communicate with a given sensor). As illustrated in Fig. 3, some sensors may needonly a few command passes and others may require more. Further, themessages/commands understood by each sensor may also vary. These differencesmake the sensor configuration process challenging. An ideal sensor configurationapproach that is designed for the IoT paradigm should be able to handle suchheterogeneity. It should also be scalable and should provide support for newsensors as they come to the market.

Fig.3 Heterogeneity in term of communication and message/command passingsequences. Some sensors may need only a few message/command passes andothers may require more. The messages/commands understood by each sensor mayalso vary

3. Scheduling, Sampling Rate, and Network Communication: The sampling ratedefines the frequency with which sensors need to generate data (i.e. sense thephenomenon) (e.g. sense temperature every 10 s). Deciding the ideal (e.g. balancebetween user requirement and energy consumption) sampling rate can be a verycomplex task and has a strong relationship with (6) Context (see below). Theschedule defines the timetable for sensing and data transmission (e.g. sense thetemperature only between 8 am and 5 pm on weekdays). Network communicationdefines the frequency of data transmission (e.g. send data to the cloud-based IoTplatform every 60 (s). Designing efficient sampling and scheduling strategies andconfiguring the sensors accordingly are challenging. Specifically, standards need tobe developed in order to define schedules that can be used across different types ofsensor devices.

10

4. Data Acquisition: Such methods can be divided into two categories: based onresponsibility and based on frequency [17]. There are two methods that can beused to acquire data from a sensor based on responsibility as illustrated in Fig. 4:push (e.g. the cloud requests data from a sensor and the sensor responds with data)and pull (e.g. the sensor pushes data to the cloud without continuous explicit cloudrequests).Further, based on frequency, there are two data acquisition methods: instant (e.g.send data to the cloud when a predefined event occurs) and interval (e.g. send datato the cloud periodically). Pros, cons, and explicabilities of these differentapproaches are discussed in [17]. Using the appropriate data acquisition methodbased on context information is essential to ensure deficiency.5. Dynamicity:This means the frequency of changing positions/appearing/disappearing of thesensors at a given location. IoT envisions that most of the objects we use ineveryday lives will have sensors attached to them in the future. Ideally, we need toconnect and configure these sensors to software platforms in order to analyze thedata they generate and so understand the environment better. We have observedseveral domains and broadly identified different levels of dynamicity based onmobility.

Fig. 4 Data can be retrieved from a sensor using both push (right side) and pull(left side) communication methods. Each method has its own advantages anddisadvantages which make them suitable for different situations

2 Sensors that move/appear/disappear at a higher frequency (e.g. RFID and otherlow-level, low-quality, less reliable, cheap sensors that will be attached toconsumables such as stationery, food packaging, etc.) can be classified as highly

11

dynamic. Sensors embedded and fitted into permanent structures (such as buildingsand air conditioning systems) can be classified as less dynamic. An ideal sensorconfiguration platform should be able to efficiently and continuously discover andre-configure sensors in order to cope with high dynamicity (Fig. 4).6. Context: Context information plays a critical role in sensor configuration in theIoT. The objective of collecting sensor data is to understand the environment betterby fusing and reasoning them. In order to accomplish this task, sensor data needsto be collected in a timely and location-sensitive manner. Each sensor needs to beconfigured by considering context information. Let us consider a scenario relatedto smart agriculture to understand why context matters in sensor configuration.Severe frosts and heat events can have a devastating effect on crops. Floweringtime is critical for cereal crops and a frost event could damage the floweringmechanism of the plant. However, the ideal sampling rate could vary depending onboth the season of the year and the time of day. For example, a higher samplingrate is necessary during the winter and the night. In contrast, lower sampling wouldbe sufficient during summer and daytime. On the other hand, some reasoningapproaches may require multiple sensor data readings. For example, a frost eventcan be detected by fusing air temperature, soil temperature, and humidity data.However, if the air temperature sensor stops sensing due to a malfunction, there isno value in sensing humidity, because frost events cannot be detected withouttemperature. In such circumstances, configuring the humidity sensor to sleep isideal until the temperature sensor is replaced and starts sensing again. Suchintelligent (re-)configuration can save energy by eliminating ineffectual sensingand network communication.

12

References:

[1]. Cristea, V., Dobre, C., Costan, A., Pop, F.: Middleware and architectures forspace-based situated computing. Int. J. Space-Based Situated Comput. 1(1), 43–58(2011)[2]. Vermesan, O., Friess, P., Guillemin, P., Gusmeroli, S., Sundmaeker, H., Bassi,A., Jubert, I.S., Mazura, M., Harrison, M., Eisenhauer, M., et al.: Internet of thingsstrategic research roadmap. In: Vermesan, O., Friess, P., Guillemin, P., Gusmeroli,S., Sundmaeker, H., Bassi, A., et al. (eds.) Internet of Things: GlobalTechnological and Societal Trends, pp. 9–52 (2011)[3]. Chen, C., Helal, S.: A device-centric approach to a safer internet of things. In:Proceedings of the 2011 International Workshop on Networking and ObjectMemories for the Internet of things, pp. 1–6. ACM (2011)[4]. Project, C.E.F.: RFID and the inclusive model for the internet of things.http://www.grifsproject.eu/data/File/CASAGRASFinalReport.pdf (2012).Accessed 15 July 2013[5]. Tan, L., Wang, N.: Future internet: The internet of things. In: 2010 3rdInternational Conference on Advanced Computer Theory and Engineering(ICACTE), vol. 5, pp. V5–376. IEEE (2010)[6]. Costa, P.D., Guizzardi, G., Almeida, J.P.A., Pires, L.F., van Sinderen, M.:Situations in con- ceptual modeling of context. In: EDOC Workshops, p. 6 (2006)[7. Dey, A.K., Abowd, G.D., Salber, D.: A conceptual framework and a toolkit forsupporting the rapid prototyping of context-aware applications. Hum. Comput.Interact. 16(2), 97–166 (2001)[8]. Ye, J., Dobson, S., McKeever, S.: Situation identification techniques inpervasive computing: a review. Pervasive Mob. Comput. 8(1), 36–66 (2012)[9]. Christensen, H.B.: Using logic programming to detect activities in pervasivehealthcare. In: Logic Programming, pp. 421–436. Springer (2002)[10]. Yau, S.S., Liu, J.: Hierarchical situation modeling and reasoning forpervasive computing. In: The Fourth IEEE Workshop on Software Technologiesfor Future Embedded and Ubiquitous Systems, 2006 and the 2006 SecondInternational Workshop on Collaborative Computing, Integration, and Assurance.SEUS 2006/WCCIA 2006, pp. 6–15. IEEE (2006)[11]. Kalyan, A., Gopalan, S., Sridhar, V.: Hybrid context model based onmultilevel situation theory and ontology for contact centers. In: Third IEEEInternational Conference on Pervasive Computing and CommunicationsWorkshops, 2005. PerCom 2005 Workshops, pp. 3–7. IEEE (2005)

13

[12]. McKeever, S., Ye, J., Coyle, L., Dobson, S.: Using dempster-shafer theory ofevidence for situation inference. In: Smart Sensing and Context, pp. 149–162.Springer (2009)[13]. Denœux, T.: A k-nearest neighbor classification rule based on dempster-shafer theory. In: Classic Works of the Dempster-Shafer Theory of BeliefFunctions, pp. 737–760. Springer (2008)[14]. Van Kasteren, T., Noulas, A., Englebienne, G., Kr ِ◌se, B.: Accurate activityrecognition in a home setting. In: Proceedings of the 10th International Conferenceon Ubiquitous Computing, pp. 1–9. ACM (2008)[15]. Doukas, C., Maglogiannis, I., Tragas, P., Liapis, D., Yovanof, G.: Patient falldetection using support vector machines. In: Artificial Intelligence and Innovations2007: From Theory to Applications, pp. 147–156. Springer (2007)[16]. Kanda, T., Glas, D.F., Shiomi, M., Ishiguro, H., Hagita, N.: Who will be thecustomer? A social robot that anticipates people’s behavior from their trajectories.In: Proceedings of the 10th International Conference on Ubiquitous Computing,pp. 380–389. ACM (2008)[17]. Perera, C., Zaslavsky, A., Christen, P., Georgakopoulos, D.: Context awarecomputing for the internet of things: a survey. IEEE Commun. Surv. Tut. 16(1),414–454 (2014). doi:10.1109/ SURV.2013.042313.00197