PaperlayerView exact revision on arXiv

Source-linked AI summary

Context Aware Computing for The Internet of Things: A Survey

Charith Perera, Arkady Zaslavsky, Peter Christen, Dimitrios Georgakopoulos

arXiv:1305.0982v1cs.SEcs.HC

TL;DR

Rapidly growing sensor deployments produce large volumes of data whose value depends on context-aware interpretation, modelling, reasoning, and distribution. The paper surveys context awareness from an IoT perspective by analysing 50 projects with a practical taxonomy, finding that context awareness remains insufficiently addressed in IoT-focused solutions and outlining future research directions.

  • Problem

    Rapidly growing sensor deployments generate data that requires context-aware interpretation, while existing IoT solutions have not adequately addressed context awareness.

  • Method

    The paper analyses and compares 50 context-aware computing projects from 2001–2011 using a practical taxonomy based on project capabilities and component combinations.

  • Results

    The survey finds that most IoT middleware solutions do not provide context-awareness functionality, despite strongly focusing on device management.

  • Takeaways & Limitations

    The survey provides a foundation for applying lessons from prior context-aware systems to IoT and identifying open challenges and future research directions.

Abstract

from arXiv · show

As we are moving towards the Internet of Things (IoT), the number of sensors deployed around the world is growing at a rapid pace. Market research has shown a significant growth of sensor deployments over the past decade and has predicted a significant increment of the growth rate in the future. These sensors continuously generate enormous amounts of data. However, in order to add value to raw sensor data we need to understand it. Collection, modelling, reasoning, and distribution of context in relation to sensor data plays critical role in this challenge. Context-aware computing has proven to be successful in understanding sensor data. In this paper, we survey context awareness from an IoT perspective. We present the necessary background by introducing the IoT paradigm and context-aware fundamentals at the beginning. Then we provide an in-depth analysis of context life cycle. We evaluate a subset of projects (50) which represent the majority of research and commercial solutions proposed in the field of context-aware computing conducted over the last decade (2001-2011) based on our own taxonomy. Finally, based on our evaluation, we highlight the lessons to be learnt from the past and some possible directions for future research. The survey addresses a broad range of techniques, methods, models, functionalities, systems, applications, and middleware solutions related to context awareness and IoT. Our goal is not only to analyse, compare and consolidate past research work but also to appreciate their findings and discuss their applicability towards the IoT.

I. INTRODUCTION

The paper frames context awareness as increasingly important for interpreting expanding sensor data in the IoT. It surveys prior work broadly, using a project-based taxonomy to assess applicability and future directions.

  • Motivation: Large-scale sensor deployments generate data whose value depends on analysis, interpretation, and contextual understanding.Context-aware computing links context information to sensor data to support more meaningful interpretation.
  • Motivation: Traditional approaches that connect sensors individually to applications become infeasible as sensor numbers and data volumes increase.This inefficiency has motivated middleware solutions addressing issues including device management, interoperability, portability, context awareness, security, and privacy.
  • Survey scope: Its practical taxonomy is built from features, models, techniques, functionalities, and approaches identified across past projects rather than imposed theoretically.The analysis focuses on higher-level capabilities instead of implementation or code-level differences.
  • Objectives: The paper aims to learn from past context-aware techniques, assess their use in IoT and other future paradigms, and identify open challenges and research directions.It also discusses how component combinations are designed, developed, and used to evaluate IoT applicability.
  • Survey scope: The survey evaluates 50 small- and large-scale projects conducted during 2001–2011 using criteria covering focus, techniques, popularity, and comprehensiveness.The reviewed projects represent research and commercial solutions in context-aware computing.

C. IoT Application Domains

The IoT connects everyday objects through sensing and communication infrastructure to support applications across industry, environment, and society. Its scale and heterogeneity create middleware and integration challenges.

  • Application domains: IoT applications span industry, environment, and society, including supply chains, healthcare, smart buildings, agriculture, recycling, and disaster alerting.One cited classification lists 54 application domains grouped into twelve categories.
  • Scale: 50 to 100 billion devices were projected to be connected to the Internet by 2020.The projection includes Internet-enabled PCs, mobile phones, and future smart objects.
  • Infrastructure: Sensor networks use wired or wireless communication among homogeneous or heterogeneous nodes, and multiple networks can connect through the Internet.IoT follows a three-layer sensor-network architecture in which multiple sensor networks connect over the Internet.
  • Infrastructure: Wireless sensor networks use technologies ranging from Bluetooth and Wi-Fi to WiMAX, cellular networks, and satellite systems.They also use both IP-based and non-IP-based communication protocols.
  • Middleware: Sensor-network middleware research addresses abstraction, data fusion, resource constraints, topology, adaptability, scalability, security, and quality of service.The surveyed approaches include IrisNet, JWebDust, Hourglass, HiFi, Cougar, Impala, SINA, Mate, TinyDB, Smart Object, Agilla, TinyCubus, TinyLime, EnviroTrack, Mires, Hood, and Smart Messages.

F. Layers in Sensor Networks

Sensor-network layers are distinguished by device capabilities, with data moving from sensor nodes through computational devices to cloud processing. The IoT extends this structure to support diverse devices, interactions, and applications.

  • F. Layers in Sensor Networks: Data flows from low- and high-end sensor nodes through sink nodes and computational devices before reaching the cloud.The cloud shares, stores, and processes the data significantly.
  • F. Layers in Sensor Networks: Six layers are identified according to processing, memory, communication, and energy capacity, which increase from layer one to layer six.Data may be processed at any layer, so management should account for capability differences.
  • G. Relationship Between Sensor Networks and IoT: The IoT comprises sensors and actuators, collects and processes sensor data, makes decisions, and performs actions through actuators.Sensor networks provide the essential sensing and actuation infrastructure within the IoT.
  • G. Relationship Between Sensor Networks and IoT: Unlike application-specific sensor networks, the IoT is a general-purpose sensor network intended to support many kinds of applications.Sensor networks can exist independently, but the IoT depends on them for much of its hardware infrastructure.
  • IoT Characteristics: IoT solutions must handle intelligence, architectural diversity, system complexity, scale, time, space, and everything-as-a-service considerations.These characteristics affect IoT solutions throughout design, development, implementation, and evaluation.
  • Middleware Support for IoT: Most existing IoT middleware solutions focus on device management, while the majority do not provide context-awareness functionality.The survey identifies this gap as motivation for evaluating context-aware solutions from other paradigms.

III. CONTEXT AWARENESS FUNDAMENTALS

Context awareness fundamentals distinguish raw sensor data from processed context and establish context as information characterizing an entity’s situation. The section reviews competing definitions and event categories used in IoT context management.

  • Scope: The section surveys context definitions, context-aware features, context types and categories, levels and characteristics, and context-management design principles for IoT.Earlier definitions are reviewed because researchers have proposed multiple formulations of context.
  • Definitions: Context is information that can be used to characterise the situation of a person, place, or object relevant to an interaction.The paper adopts Abowd et al.’s definition because it supports identifying context from data generally.
  • Raw Data and Context: Raw sensor data is unprocessed data retrieved directly from a source, whereas context is generated by processing sensor data and adding consistency checks and metadata.GPS readings become context when represented as a geographical location.
  • Events: Context may be defined as a set of interrelated events with logical and timing relations, including discrete and continuous events.Discrete events produce separate instances at sampled times, whereas continuous events persist for at least the sampling interval.

2) Definition of Context-awareness:

The paper defines context-awareness as using context to provide task-relevant information or services, and frames context management around acquisition, representation, delivery, and reaction. It distinguishes context models, context attributes, context types, and categorisation perspectives for IoT systems.

  • Context-awareness uses context to provide information or services relevant to the user’s task.
  • Context-aware frameworks typically support context acquisition, representation, delivery, and reaction.
  • Applications can use no application-level, implicit, or explicit context models, with explicit models separating context management from application code.
  • Context models specify attainable and exploitable context subsets, while context attributes identify modeled elements through identifiers, types, values, and optional properties.
  • Operational categorisation captures how context data was acquired, while conceptual categorisation captures relationships among context concepts.
  • Primary context is retrieved without existing context or sensor fusion, whereas secondary context is derived from existing context or fused data.

D. Levels of Context Awareness and characteristics

Context awareness ranges from user-configured personalisation to passive recommendations and autonomous action. IoT management frameworks must also address distributed, mobile, resource-constrained, extensible, and explainable context processing.

  • Levels of Context Awareness: Personalisation lets users manually specify preferences, while passive and active context-awareness provide options or act autonomously based on monitored situations.
  • Characteristics: IoT context is sensed by constrained devices from distributed sources, changes continuously, involves mobile objects, and is temporal, spatial, imperfect, and uncertain.
  • Design Principles: Debugging tools should expose reasoner justifications to help evaluate inaccuracies and support trustworthy IoT systems.
  • Design Principles: Frameworks should combine multiple reasoning models because no single model accommodates all IoT demands.
  • Design Principles: Mobility support requires framework variants that run across devices with different hardware and software capabilities.
  • Design Principles: Distributed IoT architectures require context sharing between frameworks and applications in both real-time and historical forms.
  • Design Principles: At IoT scale, small improvements in data structures or processing can substantially affect storage and energy consumption.
  • Design Principles: Real-time event detection is a major challenge because detected events trigger autonomous actions in IoT systems.

IV. CONTEXT LIFE CYCLE

The survey treats context management as a life cycle describing context movement from generation to consumption, including both general enterprise and specialised context approaches. Its simplest cycle has four essential phases: acquisition, modelling, reasoning, and dissemination.

  • A data life cycle explains where data is generated, how it moves between phases, and where it is consumed.
  • Context management has become an essential functionality and is offered through context services such as Context-as-a-Service.
  • Enterprise Lifecycle Approaches are robust and industry-established, whereas Context Lifecycle Approaches specialise in context but are less tested or standardised.
  • The survey analyses ten popular data life cycles, dividing the first five into enterprise approaches and the remaining five into context lifecycle approaches.
  • The simplest context life cycle comprises acquisition from physical or virtual sensors, modelling, reasoning, and dissemination of high- and low-level context.
  • Context-aware middleware must consider acquisition responsibility, frequency, source, sensor type, and acquisition process.

1) Based on Responsibility:

The survey organises context acquisition by responsibility, event frequency, source, and sensor type. These distinctions describe who initiates acquisition, when data is collected, where it comes from, and how sensor data is produced or combined.

  • Based on Responsibility: Pull acquisition requests data from sensors periodically or instantly, whereas push acquisition sends data from sensors to acquiring software.
  • Based on Frequency: Instant events occur at a point in time, while interval events span a period and require periodic sensing; both can use push or pull.
  • Based on Source: Context can be acquired directly from sensor hardware, through middleware infrastructure, or from context servers.
  • Based on Sensor Types: Physical sensors generate low-level context, virtual sensors retrieve data from external sources, and logical sensors combine physical and virtual sensors.

5) Based on Acquisition Process:

Context can be acquired by sensing, deriving information computationally, or receiving it manually. The survey also describes modelling context, organizing it into repositories, and representing it through several techniques.

  • Acquisition Process: Context is acquired by sensing data, deriving information computationally, or receiving it manually from users.Derived information may come from web-service calls or mathematical functions, while manual input includes preferences.
  • Context Modelling: Context modelling defines attributes, relationships, quality-of-context properties, and queries for synchronous requests.
  • Context Modelling: The resulting model is validated, merged into the existing context repository, and made available for later use.
  • Context Modelling: Choosing context attributes remains subjective because no standard specifies which information should be included in a model.
  • Context Life Cycle: Context information may move through ready, suspended, resumed, expired, or terminated states as its validity changes.
  • Modelling Techniques: The survey compares six popular modelling techniques: key-value, markup, graphical, object-based, logic-based, and ontology-based modelling.Key-value modelling is simple and manageable for small data volumes but cannot represent complex structures, hierarchies, or relationships effectively.

3) Graphical Modelling:

Graphical modelling represents context through relationships and can serve as persistent storage, while ontology-based approaches emphasize semantic representation, reuse, interoperability, and reasoning. The survey notes trade-offs involving complexity, retrieval, and computational cost.

  • 3) Graphical Modelling: Graphical modelling captures relationships among context elements and can be implemented with SQL, NoSQL, XML, or other representations.
  • 3) Graphical Modelling: Graphical modelling supports large-scale storage and relatively fast retrieval, but diverse implementations can hinder interoperability.
  • 3) Graphical Modelling: Complex retrieval may require difficult SQL queries, while later additions or structural changes can also be difficult.
  • 6) Ontology Based Modelling: Ontological context retrieval can become computationally intensive and time-consuming as data volume increases and structures become more complex.
  • 6) Ontology Based Modelling: Ontologies organize context semantically and provide expressive representation, reasoning mechanisms, knowledge sharing, and separation from application code.
  • 6) Ontology Based Modelling: Ontology development begins by defining the domain and scope, then reviewing existing ontologies for opportunities to reuse them.
  • 6) Ontology Based Modelling: RDF, RDFS, and OWL are ontology languages, with OWL 2 identified as the current recommendation in the survey.
  • 6) Ontology Based Modelling: Combining modelling techniques is presented as more effective than relying on any single technique, because their weaknesses can offset one another.

C. Context Reasoning Decision Models

Context reasoning deduces higher-level knowledge from available context, addressing imperfect and uncertain raw data. The survey reviews major reasoning categories and emphasizes that processing all IoT context may be infeasible.

  • Context Reasoning: Context reasoning deduces new or higher-level knowledge from available context, partly because raw context can be imperfect or uncertain.
  • Decision Models: Reasoning models include decision trees, naive Bayes, hidden Markov models, support vector machines, neural networks, Dempster-Shafer, ontology-based, rule-based, and fuzzy reasoning.
  • IoT Constraints: More than 50 billion sensors are expected to produce enormous information volumes, making the use of all context infeasible because of processing, power, and storage demands.
  • Survey Scope: The survey introduces reasoning techniques to clarify their role in IoT context awareness rather than to provide a comprehensive reasoning-technique survey.
  • Context Reasoning: The survey classifies reasoning into supervised learning, unsupervised learning, rules, fuzzy logic, ontological reasoning, and probabilistic reasoning.
  • Supervised Learning: Supervised learning derives a function from labeled training examples and is used in mobile-phone sensing and activity recognition.
  • Unsupervised Learning: Unsupervised techniques find hidden structures in unlabeled data and support tasks ranging from sensor-network routing to positioning and user-context profiling.

3) Rules:

Rules provide straightforward high-level inference from low-level context, while other reasoning approaches address graded truth, semantics, uncertainty, and evidence combination. The survey situates these techniques within context distribution and project evaluation.

  • 3) Rules: Rule-based reasoning commonly uses IF-THEN-ELSE structures to generate high-level context from low-level context.The survey identifies rules as the most popular reasoning method according to Figure 7.
  • Fuzzy Logic: Fuzzy logic permits partial truth and degrees of membership, allowing approximate reasoning for facts that are not crisp.
  • Ontology-Based Reasoning: Ontological reasoning uses RDF(S) and OWL(2), integrates with ontology modelling, and can be complemented by rules such as SWRL.
  • Probabilistic Logic: Probabilistic techniques attach probabilities to facts, combine sensor evidence, resolve conflicts, and help identify event occurrences.
  • Hybrid Reasoning: Because individual techniques have different strengths and weaknesses, hybrid approaches combine models such as logic rules, Bayesian networks, and CoCoGraphs.
  • Context Distribution: Context consumers receive results either through queries or subscriptions that return data periodically or when specified events occur.
  • Evaluation Framework: The evaluation framework compares 50 projects using a taxonomy covering project focus, chronology, citations, and documented functionality.

B. Evaluation of Research Efforts

The evaluated systems use diverse architectures and context-management components to collect, model, reason over, store, and distribute context across application domains. Together, they illustrate middleware approaches for mobile, ubiquitous, smart-environment, and IoT settings.

  • Development frameworks: Context Toolkit supports application development through context widgets, interpreters, and additional abstractions for retrieving and reasoning over sensor data.It is described as an early framework for developing and deploying context-aware applications.
  • Distributed and mobile systems: Aura uses distributed task, environment, observer, and supplier components to preserve work continuity across environments and adapt to changing resources.Its architecture targets multiple computational devices used by people in everyday settings.
  • Ontology-based middleware: Ontology-based systems such as CoBrA, SOCAM, Hydra, and Feel@Home represent context for reasoning, device integration, layered domain modelling, or cross-domain interaction.Their designs address smart spaces, IoT middleware, domain-specific knowledge, and interactions across smart-home, office, and mobile domains.
  • Context management: Context-management architectures combine acquisition, processing, modelling, storage, selection, filtering, and reasoning components to handle sensor data and context.HCoM combines semantic ontologies with relational schemas, while COSMOS separates collection, processing, and adaptation into layers.
  • Specialized approaches: Other evaluated solutions emphasize mobile context gathering, agent-based sensor aggregation, event-driven rule checking, service-based management, and adaptive middleware.These approaches include CaSP, SIM, DMS-CA, MoCA, and CARISMA, each targeting different context-management or adaptation requirements.

1) Development Aids and Practices:

The survey identifies development practices that support IoT-scale context systems, including standard interfaces, extensible models, privacy mechanisms, non-technical deployment, and operational metadata.

  • Development aids and practices: Standard interfaces improve interoperability and extendibility by allowing context components and techniques to interact despite differing internal implementations.The survey presents standardisation as especially important when new components must be added to existing IoT systems.
  • Development aids and practices: Data-representation choices involve trade-offs: binary formats are compact and portable, objects support complex structures, and attribute-value pairs offer more limited complexity.The survey frames modelling and communication choices as critical IoT design decisions with different advantages and disadvantages.
  • Development aids and practices: CoOL demonstrates how context-management extensions such as modelling and reasoning can be added to general-purpose service models.An access point handles communication between CoOL and the surrounding general-purpose architecture.
  • Development aids and practices: Security and privacy remain insufficiently considered in many context-aware solutions, while CoBrA illustrates policy-based privacy management with ontologies.The cited approach uses policies to control monitoring and access to contextual information.
  • Development aids and practices: Octopus highlights programming models for sensor deployment by non-technical users, while operational metadata captures complexity, quality, freshness, and acquisition cost.These practices address deployment usability and the operational characteristics of context data.

2) Mobility, Validity, and Sharing:

The survey emphasizes that IoT context systems must remain mobile, valid, shareable, extensible, energy-aware, dynamically configurable, and capable of distributed real-time processing.

  • Mobility, Validity, and Sharing: Mobility requires middleware to preserve context-aware functionality across situations, devices, platforms, and resource constraints.Aura is cited as an example spanning cloud servers, computers, tablets, phones, and everyday objects.
  • Mobility, Validity, and Sharing: Validation, conflict resolution, and context sharing support more accurate reasoning and smoother transitions between devices or situations.The survey connects these functions with competing data sources, consistency checking, privacy concerns, and higher-level context recognition.
  • On Demand Data Modelling: Extensible ontology models separate general-purpose and domain-specific knowledge so IoT systems can expand across unpredictable domains.SOCAM allows upper and domain-specific ontologies to be extended independently.
  • On Demand Data Modelling: Multiple reasoning techniques can complement one another, with statistical and ontological reasoning combined in COSAR to achieve more accurate results.The survey also identifies Bayesian, probabilistic, and fuzzy approaches as useful in different situations.
  • Hardware Layer Support: Embedding context awareness in hardware-layer communication can make sensors act more intelligently and save energy in machine-to-machine environments.The survey argues that fully context-aware solutions should incorporate capabilities in both software and hardware layers.
  • Dynamic Configuration and Extensions: Pluggable rules, automated sensor connection, and dynamic composition enable configuration and extension when domains, devices, and interactions cannot be predicted in advance.Hydra, ACoMS, UPnP, Solar, CMF, COSMOS, and COPAL illustrate complementary mechanisms for extensibility and usability.
  • Distributed Processing: Real-time queries and stream processing must combine live and historical context while supporting cross-domain queries across distributed managers.The survey identifies real-time event detection and distributed context sharing as essential IoT processing requirements.

8) Other Aspects:

The survey identifies IoT-specific challenges in automating sensor connections, interpreting and selecting sensor data, integrating techniques and middleware, and protecting context across layers. It uses fifty projects and past research to frame these challenges and future directions.

  • Challenges and future research directions: Fifty context-aware projects were evaluated to identify lessons from past research and six challenges requiring novel IoT techniques.The survey examines context-aware computing from an IoT perspective and discusses future research directions.
  • Sensor connection and configuration: Billions of Internet-connected sensors make manual sensor-to-application or middleware configuration infeasible, requiring automated or semi-automated connection methods.This challenge differs from traditional pervasive and ubiquitous systems, which connect only limited numbers of sensors.
  • Context discovery: Automatically understanding and annotating sensor data across varied application domains remains challenging, with semantic technologies and linked data identified as future directions.Semantic technology is described as a way to encode domain knowledge.
  • Technique integration: No single acquisition, modelling, or reasoning technique meets IoT requirements, while integrating multiple techniques has shown promising success.The survey notes that the immaturity of IoT makes it difficult to predict when and where each technique should be used.
  • Sensor selection: IoT sensor selection must address both which sensors provide information about a target phenomenon and which sensor to use when multiple alternatives measure the same parameter.The survey illustrates these questions with environmental pollution measurement and pH concentration sensing.
  • Security, privacy, and trust: Security and privacy protections must span hardware, communication, annotation and discovery, modelling, and distribution layers to support user trust.The survey identifies middleware-level protection as largely unattended in existing context-aware solutions.
Loading 1305.0982v1…