PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey of Data Fusion in Smart City Applications

Billy Pik Lik Lau, Sumudu Hasala Marakkalage, Yuren Zhou, Naveed Ul Hassan, Chau Yuen, Meng Zhang, U-Xuan Tan

arXiv:1905.11933v1eess.SP

TL;DR

Smart city applications increasingly combine diverse data sources, creating a need for systematic evaluation of data fusion. This paper proposes and applies a six-perspective classification across selected application domains, then discusses trends, challenges, and deployment practices.

  • Problem

    The expanding scale and scope of smart city data sources, collection techniques, and processing architectures require evaluation across multiple aspects.

  • Method

    The paper proposes a classification with six perspectives and applies it to selected data fusion applications across smart city domains.

  • Results

    The paper provides an overall view of data fusion techniques and current trends across selected smart city application domains.

  • Takeaways & Limitations

    The classification helps readers grasp data fusion trends and supports understanding and improvement of smart city domains.

Abstract

from arXiv · show

The advancement of various research sectors such as Internet of Things (IoT), Machine Learning, Data Mining, Big Data, and Communication Technology has shed some light in transforming an urban city integrating the aforementioned techniques to a commonly known term - Smart City. With the emergence of smart city, plethora of data sources have been made available for wide variety of applications. The common technique for handling multiple data sources is data fusion, where it improves data output quality or extracts knowledge from the raw data. In order to cater evergrowing highly complicated applications, studies in smart city have to utilize data from various sources and evaluate their performance based on multiple aspects. To this end, we introduce a multi-perspectives classification of the data fusion to evaluate the smart city applications. Moreover, we applied the proposed multi-perspectives classification to evaluate selected applications in each domain of the smart city. We conclude the paper by discussing potential future direction and challenges of data fusion integration.

I. INTRODUCTION

The paper motivates data fusion for increasingly complex smart city applications and proposes a multi-perspectives classification to evaluate them across domains and challenges.

  • Motivation: Growing urban populations and expanding ICT, IoT, Big Data, data mining, and data fusion motivate smart city development.The paper connects these technologies with managing urban resources and infrastructure for sustainable living.
  • Motivation: Smart city applications draw on diverse communication networks and increasingly abundant data sources.The paper identifies 5G, WSN, LPWAN, and NB-IoT as communication technologies supporting data transfer and sensing.
  • Research gap: The paper argues that expanding data sources, collection methods, and processing architectures require evaluation from multiple perspectives.This need arises in highly complicated applications that use data from various sources and assess performance across multiple aspects.
  • Contribution: The proposed classification evaluates data fusion through objectives, techniques, input and output types, source types, scales, and platform architectures.These perspectives are applied across seven smart city domains, including Smart Living, Smart Environment, and Smart Infrastructure.
  • Contribution: The paper reviews selected notable applications, identifies data fusion trends and challenges, and outlines practices for smart city deployment.The authors explicitly state that this is not a comprehensive review of all smart city applications.

II. DATA FUSION CLASSIFICATION USING MULTI-PERSPECTIVES

The classification organizes smart city data fusion into six perspectives and 30 classes, then defines four application objectives that motivate fusion.

  • Framework: The framework draws on prior smart city reviews and established non-smart-city classifications, including Dasarathy, Whyte, JDL, and architecture-based approaches.These prior schemes emphasize input/output types, data relationships, processing levels, or system architecture.
  • Framework: The proposed classification comprises six perspectives: objectives, techniques, data input and output types, source types, system scales, and platform architectures.The framework contains 30 classes, with reference codes such as O1 and S3.
  • Data Fusion Objectives (O): The four fusion objectives are fixing problematic data, improving data reliability, extracting higher-level information, and increasing data completeness.They address quality issues, noise and redundancy, knowledge extraction from raw sources, and coverage limitations.
  • Data Fusion Objectives (O): Problematic-data fusion addresses inconsistency, imperfection, and disparateness in data sources.The paper classifies this objective as O1 and presents it as an approach for overcoming data-quality problems.
  • Data Fusion Objectives (O): Reliability-oriented fusion adds redundant sources in noisy or less controlled environments, while completeness-oriented fusion combines sources to overcome limited coverage.The paper also places security enhancement under the Data Reliability class.
  • Data Fusion Objectives (O): Higher-level information extraction combines raw sources to infer information that cannot be directly obtained from individual measurements.The paper illustrates this with detecting building occupancy from multiple ambient sensors.

B. Data Fusion Techniques (T)

The paper groups data fusion techniques into association, estimation, decision fusion, machine learning, dimensionality reduction, statistical analysis, and visualization approaches.

  • Core Fusion Techniques: Data association fuses sources according to similarity, while state estimation uses multiple sources to improve estimation accuracy.Examples include nearest neighbors, probabilistic data association, maximum likelihood, Kalman filters, and particle filters.
  • Core Fusion Techniques: Decision fusion combines decisions from system sub-components to achieve an overall objective.The paper gives robot actuation as an example and lists Bayesian and Dempster-Shafer inference among representative methods.
  • Learning-Based Techniques: Classification groups objects by characteristics, while prediction forecasts outputs from one or more data sources.Prediction includes regression and more complex forecast modeling.
  • Learning-Based Techniques: Unsupervised machine learning discovers knowledge without relying on labels, including clustering and anomaly detection.Semi-supervised learning is also categorized under this class in the paper.
  • Analytical Techniques: Dimension reduction extracts features or supports visualization while preserving data characteristics and reducing high-dimensional processing complexity.Principal Component Analysis is listed as an example.
  • Analytical Techniques: Statistical inference and analysis outlines information and common knowledge or hypotheses from input sources, while visualization presents outputs to end users.Visualization often requires human intervention.

C. Data Input and Output Types (D)

The classification organizes data fusion by the relationship between input and output data, and separately categorizes the source types used in smart city applications.

  • DAI-DAO fuses multiple raw data sources to increase reliability while producing raw-data output.
  • DAI-FEO combines raw sources to extract features describing an observed system for further extraction or decision-making.
  • FEI-FEO combines features from different sensors to generate new features and is commonly called feature fusion.
  • FEI-DEO fuses system features to make decisions, including actuation of system components.
  • DEI-DEO combines decisions from different sources to obtain a final output decision.
  • Source categories include physical, cyber, participatory, and hybrid data, with hybrid sources mixing sources such as participatory and physical sensor data.

E. Data Fusion Scales (L)

Data fusion scales are defined by sensor coverage, ranging from individual sensors to multiple cities or larger terrains; platform architectures span edge, fog, cloud, and hybrid layers.

  • Data Fusion Scales (L): Fusion scale is based on sensor coverage and includes sensor, building-wide, inter-building, city-wide, and inter-city-or-larger classes.
  • Data Fusion Scales (L): Sensor-level fusion combines data from multiple physical sensors, including various sensors in a smartphone.
  • Data Fusion Scales (L): Building-wide fusion combines data collected within one premise, while inter-building fusion combines sources across several buildings in a small area.
  • Data Fusion Scales (L): City-wide fusion uses sources covering an entire city, whereas inter-city fusion covers one or more cities or larger terrains.
  • Platform Architectures (P): Edge processing occurs near collection sites and can significantly reduce communication overheads and latency; fog processing occurs between edge and cloud.
  • Platform Architectures (P): Cloud platforms process and fuse data in the cloud, providing online and offline access but increasing communication overheads and costs.
  • Platform Architectures (P): Hybrid platforms distribute processing across at least two layers, using edge or fog for low-level fusion and cloud for high-level information extraction.

III. SMART CITY APPLICATIONS OVERVIEW

The overview applies multi-perspectives classification to selected data-fusion applications across smart city domains and discusses their data sources, techniques, and research trends.

  • The evaluation uses a generic multi-perspectives classification because smart city applications have diverse requirements and techniques.
  • Smart Living: Smart living targets urban citizens’ liveability, with applications seeking higher-level information or greater data completeness and often using cloud or hybrid platforms.
  • Smart Living: Healthcare applications use technologies such as smartphones and sensing devices, including electroencephalographic signals and voice for telehealth monitoring.
  • Smart Living: Smart homes provide control and monitoring of appliances and services through ICT implementation.
  • Smart Living: Smart community applications can fuse information from multiple sources to monitor behaviour and generate a single credit score for each person.
  • Smart Living: The paper identifies privacy as a debated issue associated with such data-state practices.

B. Smart Urban Area Management

Smart urban area management uses ICT across urban planning, governance, and smart buildings, commonly pursuing higher-level information or greater data completeness.

  • Smart urban area management covers urban planning, governance, and smart buildings, with applications operating at least at building scale.
  • Data fusion in this domain commonly extracts higher-level information or increases data completeness.
  • Smart Governance: Governance integration is challenging because transparent services must be balanced with protection of sensitive information, leaving limited study materials.
  • Urban Planning: Traditional urban planning combines aerial photography with statistics, but frequently lacks fine detail and produces unrepresentative outputs.
  • Urban Planning: Combining satellite and aerial images addresses missing detail in existing urban structures, while low-power sensors offer broader coverage at lower deployment cost.
  • Smart Buildings: Building management fuses parameters to predict occupancy and control HVAC while optimizing resources such as hot water, electricity, and ventilation.
  • Smart Environment: Smart environment studies internal and external surroundings, commonly using physical or hybrid sources over large spatial coverage.

1) Landscape Monitoring:

Landscape monitoring faces limited sensing coverage, addressed through mobile sensing and satellite-based data, while smart industry and agriculture apply fusion to support production, maintenance, and food resources.

  • Landscape Monitoring: Mobile sensing leverages moving vehicles or humans to provide large spatial coverage for landscape monitoring.It is not suitable for real-time applications unless multiple sensing resources are available.
  • Landscape Monitoring: Satellite images are common for modeling urban natural resources but require data enhancement before use.Earlier data-fusion work focused on improving satellite-image quality.
  • Smart Industry: Smart industry applications mainly use physical sensors, with fusion often performed at sensor or building level across manufacturing, maintenance, and agriculture.The paper divides smart industry into Smart Manufacturing, Smart Maintenance, and Smart Agriculture.
  • Smart Maintenance: Preventive maintenance uses data fusion to predict a machine’s remaining useful life so maintenance can occur on time and costs can be reduced when needed.Common prediction models include convolutional and deep neural networks using machine states, sensor readings, and related parameters.

3) Smart Agriculture:

Smart agriculture addresses urban food demand, while smart economics covers commercial activities including commerce, supply chains, and tourism supported by diverse fused data sources.

  • Smart Agriculture: Smart farming has emerged to meet smart-city food-supply demand through land and sea agriculture.Controlled-environment crop production can provide fresh supplies, although plant disease threatens dense plantations.
  • Smart Economics: Smart economics encompasses urban commercial activities in supply chains, logistics, finance, and tourism.The paper organizes it into Smart Commerce, Smart Supply Chain, and Smart Tourism.
  • Smart Commerce: E-commerce recommender systems fuse mobility, credit-card purchases, and social-media interactions to understand customers and recommend products.Customer behavior and market-research data are also fused to obtain a more holistic view.
  • Smart Supply Chain: A food-supply-chain fusion framework targets faster data processing, shelf-life prediction, and real-time supply-chain planning.The framework uses three information-fusion tiers for these objectives.
  • Smart Tourism: Tourism recommendation and analysis systems use user information, feedback, geo-tagged photos, and smartphone location data to model travel preferences and behavior.Rocchio’s method is used to fine-tune one recommendation system from user choices.

F. Smart Human Mobility

Smart human mobility fuses movement and environmental data to support positioning, mobility analysis, and transportation services, while newer systems address congestion, last-mile travel, and autonomous driving.

  • Human Mobility Understanding: Smart human mobility collects, manages, and analyzes diverse mobility data to improve understanding of transportation systems.Advanced ICT has expanded the availability of mobility-related data for researchers.
  • Human Mobility Understanding: Indoor positioning research focuses on improving accuracy, deployment cost, and energy cost because outdoor positioning and services are comparatively developed.GPS remains the dominant outdoor-positioning technology.
  • Human Mobility Understanding: Fusing movement trajectories with GIS or floor-plan data reveals mobility patterns such as origin-destination matrices, transport modes, and building flows.These results provide clues for transportation-system improvement, urban planning, and communication networks.
  • Smart Transportation Systems: Transportation-system improvement targets congestion relief, better public transportation, and new transport systems.Intelligent light control, network and schedule optimization, and big-data modeling are discussed across these goals.
  • Smart Transportation Systems: Bike-sharing analysis evaluates operational strategies, while autonomous vehicles fuse vehicle and road data through machine-learning and control algorithms.Security is important for reliable autonomous-vehicle deployment.

1) Smart Grid:

Smart infrastructure uses data fusion across electricity, energy, facilities, and communication, with research directions emphasizing data quality, representation, privacy, security, and fusion techniques.

  • Smart Grid: Smart grids integrate ICT to provide reliable, stable electricity and balance loads and demand across areas, buildings, or households.Load and demand balancing is the common goal of this sub-domain.
  • Smart Energy: Renewable-energy research seeks greener, lower-carbon energy production, especially through solar and wind power.Solar harvesting is constrained by limited energy harvesting capacity.
  • Smart Facility: Smart facilities provide public services such as parking and water supply, with water-treatment reliability requiring attention to leakage and downtime.One cited approach combines district water-meter data for leakage detection.
  • Smart Communication: Urban communication infrastructure uses varied standards and protocols, including 5G and wireless sensor networks, to meet different application requirements.Signal fusion across multiple antennas can reconstruct transmitted information.
  • Challenges and Open Research Directions: Four open research directions are data quality, data representation, data privacy and security, and data-fusion technique.The paper presents these directions after reviewing smart-city applications that use data fusion.
  • Challenges and Open Research Directions: Data-source quality directly determines output quality because data-fusion processing follows the garbage-in-and-garbage-out principle.The paper highlights sensing coverage and sensing longevity as data-source improvement aspects.

1) Sensing Coverage:

Sensing coverage, collection mobility, sensor longevity, data representation, and communication choices constrain smart-city data fusion. The paper describes crowdsensing and mobile sensing trade-offs alongside energy-aware sensing and interoperability requirements.

  • Sensing Coverage: Insufficient sensing coverage can produce unrepresentative results while increasing hardware deployment costs and complicating communication-architecture design.The paper identifies crowdsensing and mobile sensing as approaches for addressing coverage limitations.
  • Sensing Coverage: Crowdsensing uses personal mobile devices cost-efficiently, but privacy intrusion, battery consumption, geolocation gaps, and randomly distributed data can reduce data quality.Fixed-time and fixed-location collection with participant incentives is suggested for filtering invalid information.
  • Sensing Coverage: Mobile sensing follows designated vehicle routes, expanding urban collection but limiting spatial resolution in pedestrian paths and residential areas.Combining mobile sensing with crowdsensing is proposed as a workaround for broader coverage.
  • Data Sensing Longevity: Long-term collection improves temporal resolution, while energy harvesting and low-energy devices support self-sustaining sensing.Energy harvesting can draw power from solar energy, vibration, or temperature differences.
  • Data Sensing Longevity: Solar harvesting is widely available but varies with irradiance, making energy management, battery capacity, and panel efficiency important design factors.Temperature- and vibration-based harvesting remain limited to certain use cases.
  • Data Representation: Heterogeneous data formats hinder integration; ontologies and semantic-web standards such as RDF support interoperable data exchange, although many ontology languages remain domain-specific.Domain-specific ontologies can contribute to segmentation between smart-city domains.

C. Privacy and Security

Smart-city data fusion must address privacy risks from sensitive resident data and security risks across infrastructure and data handling. The paper highlights consent, encryption, architecture, standards, and implementation practices as central concerns.

  • Privacy: Collecting residents’ data is difficult because poorly managed sensitive information can be misused for information theft or identity fraud.The paper identifies privacy protection as a major challenge in smart-city applications.
  • Privacy: Privacy regulations emphasize user consent, but inadequate storage security can compromise collected data and damage reputation and public trust.The Facebook and Cambridge Analytica scandal is cited as an example of potential data misuse.
  • Security: Smart-city security concerns cover both technology and infrastructure, including data centers and architectures, and data security across generation, storage, and communication.The paper attributes this two-part framing to Kitchin.
  • Security: System-architecture security depends on deployment design, ranging from traditional client-server to decentralized architectures and incorporating security standards and practices.The paper describes continuous security enhancement as technology changes rapidly.
  • Security: Encryption restricts data access to authorized parties, while attribute-based encryption provides fine-grained access control, scalable key management, and flexible data distribution.These mechanisms address data security during generation, storage, and communication.
  • Security: Despite ongoing security research, increasing breaches and cyber threats are linked in the paper to negligence in security practices and implementation.The paper argues that security is often treated as an afterthought and recommends compliance with security standards.

1) Explainable Deep Neural Network:

The paper positions explainable deep learning as a response to black-box modeling in smart-city data fusion, while also addressing limited labels and heterogeneous data. It presents multi-perspective classification and open directions that include explainability, representation, privacy, and fusion techniques.

  • Explainable Deep Neural Network: Machine learning is increasingly used for high-dimensional smart-city data, with current research emphasizing explainability for neural-network models.Explainable AI seeks semantic meaning behind modeling logic rather than treating models as black boxes.
  • Explainable Deep Neural Network: Cognitive layers can mediate between learning and explanation by converting learned abstractions, policies, and clusters into explainable formats.Interpretable models such as Bayesian learning can represent uncertainties in deep-learning development.
  • Explainable Deep Neural Network: Ground-truth collection and annotation are challenging in smart-city applications, although deep neural networks still require labels to extract higher-level information.The paper identifies unlabeled data as a problem for achieving application objectives.
  • Explainable Deep Neural Network: Transforming unlabeled data into useful features can help, but preprocessing still requires cleansing and imputation, while aggressive filtering may remove knowledge.The paper contrasts feature transformation and filtering as approaches to unlabeled or problematic data.
  • Explainable Deep Neural Network: Hybrid models combine high-level, low-level, or both types of data sources to generate domain-specific insights and address data-privacy problems alongside machine learning.An urban-planning example combines environmental sensors, urban-area feedback, and cyber data.
  • Conclusion: The proposed classification evaluates smart-city data fusion through six perspectives and applies them to selected works across application domains.The perspectives cover objectives, techniques, input and output types, source types, fusion scales, and system architecture.
  • Conclusion: The paper identifies four open research directions: data quality, data representation, data privacy and security, and data fusion technique.These directions frame future improvements to smart-city data-fusion integration.
Loading 1905.11933v1…