Machine Learning for Indoor Localization and Navigation

Preface
Acknowledgments
Contents
Part I Introduction to Indoor Localization and Navigation
An Overview of Indoor Localization Techniques
1 Introduction
2 Smartphone: Sensors and Radios
3 Indoor Localization Methods
4 Fingerprinting-Based Indoor Localization
5 Hybrid Localization Methods
6 Domain Challenges
7 Conclusions and Future Directions
References
Smart Device-Based PDR Methods for Indoor Localization
1 Introduction
2 Smart Devices and Built-in Sensors
2.1 Advantages and Carrying Modes
2.2 Sensors
2.3 Comparison of Device Usage and Methodology Complexity
3 Smart Device-Based PDR Methods
3.1 General Steps in Smart Device-Based PDR Methods
3.2 Performance Evaluation Metrics
3.3 Comparison of Techniques and Performance Evaluations
4 Challenges
4.1 Initialization
4.2 Fixed Position
4.3 Standardized Activities Classification
4.4 Battery Consumption
4.5 Long-Term Walking Trajectory
4.6 Multi-floor Localizations
4.7 Demand on External Hardware
5 Conclusion
References
Geometric Indoor Radiolocation: History, Trends and Open Issues
1 Introduction
1.1 Definitions and Taxonomy
1.2 Actors of the Localization Process
1.3 Radiolocation Architectures
1.4 Radiolocation Techniques
1.5 Metrics
2 Geometric Localization Approaches
2.1 Angle of Arrival
2.2 Phase of Arrival and Phase Difference of Arrival
2.3 Time of Arrival and Time Difference of Arrival
2.4 Received Signal Strength
2.5 Hybrid Approaches
2.6 Comments
3 The Indoor Propagation Environment
4 The Role of Machine Learning in Indoor Radiolocation
5 Summary
References
Indoor Localization Using Trilateration and Location Fingerprinting Methods
1 Introduction
2 Trilateration
2.1 Path Loss Model
2.2 Trilateration Algorithm
2.3 Filtering Techniques
3 Fingerprinting Algorithm
4 A BLE-Based Indoor Localization System
4.1 Experimental Setup and Data Collection
4.1.1 Line-of-Sight Scenario Versus Non-Line-of-Sight Scenario
4.1.2 Indoor Localization Using Fingerprinting
5 Results
5.1 LOS Scenario and NLOS Scenario Results
5.1.1 LOS Scenario
5.1.2 NLOS Scenario
5.2 Trilateration-Based Method Results
5.3 Fingerprinting-Based Method Results
5.3.1 Grid-Based Fingerprinting Positioning Results
5.3.2 Location of Interest (LOI)-Based Fingerprinting Positioning Results
6 Conclusions
References
Localization with Wi-Fi Ranging and Built-in Sensors: Self-Learning Techniques
1 Introduction
2 Localization Basics
2.1 Assumptions
2.2 Wi-Fi Ranging
2.3 Pedestrian Dead Reckoning
2.4 Kalman Filter-Based Localization
3 Machine Learning for Enhanced Wi-Fi Ranging
3.1 Measurement Campaign
3.2 Enhanced RSS-Based Ranging Using CSI
3.3 Enhanced RTT-Based Ranging
4 Self-Learning Techniques
4.1 Overview
4.2 Self-Learning Using Wi-Fi Data
4.3 Sensor-Aided Learning Technique
5 Real-World Deployment Examples
5.1 RSS-Based Localization Results
5.2 RTT-Based Localization Results
6 Conclusions
Appendix: Proof of Theorem 1
References
Part II Advanced Pattern-Matching Techniques for Indoor Localization and Navigation
Fusion of WiFi and IMU Using Swarm Optimization for Indoor Localization
1 Introduction
2 Review
2.1 Basics of WiFi RSS
2.2 Machine Learning Methods in WiFi Localization
2.2.1 Matching Models
2.2.2 Typical Regression Models
2.2.3 CNN Models
2.2.4 RNN Models
2.2.5 Simple Performance Comparison
2.3 Recent Advances
2.3.1 RSS Fingerprinting Methods
2.3.2 Filtering Methods
3 PSO-Based Localization Algorithm
3.1 Displacement Estimation
3.1.1 Step Detection
3.1.2 Step Length Estimation
3.1.3 Walking Direction Estimation
3.2 WKNN Method
3.3 Fitness Evaluation
3.4 Improved Particle Swarm Optimization
4 Experiment and Evaluation
4.1 Experimental Setup
4.2 Performance Evaluation
5 Conclusion
References
A Scalable Framework for Indoor Localization Using Convolutional Neural Networks
1 Introduction
2 Related Works
3 Convolutional Neural Networks
4 CNNLOC Framework: Overview
4.1 Overview
4.2 Pre-Processing of RSSI Data
4.3 RSSI Image Database
4.4 Hyperparameters
4.5 Integrating Hierarchy for Scalability
5 Experiments
5.1 Experimental Setup
5.2 Smartphone Implementation
5.3 Experimental Results
5.3.1 Indoor Localization Accuracy Comparison
5.3.2 CNNLOC Scalability Analysis
5.3.3 Accuracy Analysis with Other Approaches
6 Conclusions
References
Learning Indoor Area Localization: The Trade-Off Between Expressiveness and Reliability
1 Introduction
2 Related Work
2.1 Deep Learning for Fingerprinting
2.1.1 Device Perspective
2.1.2 Fingerprint Constructions
2.1.3 Models for Solving the Fingerprinting Problem
2.2 Area Localization/Zone Detection
2.2.1 RSS and Floor Plan-Independent Pre-segmentation
2.2.2 Floor Plan-Aware Pre-segmentation
2.2.3 RSS-Aware Pre-segmentation
2.2.4 Without Pre-determined Segmentation
2.3 Quantification of (Area) Localization Quality
3 Metric for Area Localization: Trade-Off Between Correctness and Expressiveness
3.1 Area Localization Score (ALS)
3.1.1 Analysis
3.1.2 Interpretation
3.1.3 Application on Point Localization Models
4 Segmentation-Based Area Classification
4.1 Data-Aware Floor Plan Segmentation
5 Segmentation-Free Area Localization
5.1 DeepLocBox: Area Localization via Bounding Box Regression
5.2 DeepLocBIM: Learning Area Localization Guided by Building Model
6 Comparison of Models
6.1 Dataset
6.2 Evaluation
7 Conclusion
References
Exploiting Fingerprint Correlation for Fingerprint-Based Indoor Localization: A Deep Learning-Based Approach
1 Introduction
2 A Cramer-Rao Lower Bound on Location Error
2.1 The CRLB for Hybrid Multiple Measurements Fingerprint Localization
2.2 The CRLB for RSS-Only and CSI-Only Fingerprint Localization
2.3 The CRLB for Hybrid RSS/CSI Localization
3 Design RSS Fingerprint-Based Indoor Localization Algorithm
3.1 Radio Map Construction
3.2 RP Clustering: Spectral Clustering
3.3 Backpropagation Neural Network Training
3.4 Online Localization
3.5 Experimental Results
4 Hybrid RSS/CSI Fingerprint for Indoor Localization Algorithm
4.1 Data Processing
4.2 Deep Auto-encoder Training
4.3 Fingerprint Database Construction
4.4 Deep Neural Network Training
4.5 Online Localization
4.6 The Construction of Fingerprint Database
4.7 Detailed Algorithm of Indoor Localization System
4.8 Experimental Results
5 Conclusion
Appendix
Proof for Corollary 1
Proof for Corollary 2
Proof for Lemma 3
Proof for Corollary 3
Proof for Corollary 4
References
On the Application of Graph Neural Networks for Indoor Positioning Systems
1 Introduction
2 Related Work
3 GNNs for Indoor Localization
3.1 Problem Statement
3.2 Graph Neural Networks
3.3 Proposed Architectures
3.3.1 Homogeneous Graphs
3.3.2 Heterogeneous Graphs
4 Implementation and Experimental Results
4.1 Graph Construction
4.2 Datasets
4.2.1 UJIIndoorLoc
4.2.2 MNAV
4.3 Experimental Results
4.3.1 On the Size of the Training Set
4.3.2 Propagation Environment Changes
4.3.3 AP Failures
5 Conclusions
References
Part III Machine Learning Approaches for Resilience to Device Heterogeneity
Overview of Approaches for Device Heterogeneity Management During Indoor Localization
1 Device Heterogeneity in Indoor Positioning
2 Adjustment Method Based on Linear Transformation
3 Calibration-Free Function Mapping Method
4 Non-absolute Fingerprint Method
5 Conclusion
References
Deep Learning for Resilience to Device Heterogeneity in Cellular-Based Localization
1 Introduction
2 Problem Statement
3 Literature Review
3.1 Cellular-Based Techniques
3.2 Deep Learning-Based Techniques
3.3 Heterogeneity Handling Techniques
4 System Overview
5 The Details of the Device-Invariant System
5.1 The Preprocessor Module
5.2 The Feature Extractor Module
5.2.1 Difference Calculator
5.2.2 Encoder Creator
5.2.3 Feature Aggregator
5.3 The Localization Model Constructor
5.3.1 The Network Architecture
5.3.2 Preventing Overfitting
5.4 Online Phase
6 Evaluation
6.1 Collection Setup and Tools
6.2 Performance Evaluation
6.2.1 Performance of the Feature Extractor Module
6.2.2 Resilience to Device Diversity
6.2.3 Performance in Different Modes of Feature Extraction
6.2.4 Different Providers
6.3 Comparative Evaluation
6.3.1 Localization Accuracy
6.3.2 Energy Consumption
7 Summary
References
A Portable Indoor Localization Framework for Smartphone Heterogeneity Resilience
1 Introduction
2 Background and Related Work
3 Heterogenous Fingerprint Analysis
4 Hidden Markov Model (HMM) Formulation
5 SHERPA Framework
5.1 Wi-Fi Fingerprinting
5.2 Fingerprint Database Pre-processing
5.3 SHERPA Offline/Training Phase
5.4 SHERPA Online/Testing Phase
5.4.1 Motion-Aware Prediction Deferral
5.4.2 Noise Resilient Fingerprint Sampling
5.4.3 Smart Noise Reduction with Boosted Scans per Prediction
5.4.4 Heterogeneity Resilient Pattern Matching: PCC
5.4.5 Z-Score-Based Reference Point Selection
5.4.6 Shape Similarity Focused Hidden Markov Model
5.4.7 Optimizing Emission Matrix for Prediction Time
6 Experimental Setup
6.1 Heterogeneous Devices and Fingerprinting
6.2 Indoor Paths for Localization Benchmarking
6.3 Comparison with Prior Work
7 Experimental Results
7.1 Performance of Localization Techniques
7.2 Sensitivity Analysis on Scans per Prediction
7.3 Sensitivity Analysis on Scan Memory
7.4 Comparison of Execution Times
8 Conclusion
References
Smartphone Invariant Indoor Localization Using Multi-head Attention Neural Network
1 Introduction
2 Recent Improvements in Fingerprinting-Based Indoor Localization
3 The Attention Mechanism
4 Analyzing RSSI Fingerprints
5 The ANVIL Framework
5.1 Data Augmentation
5.2 The Offline Phase
5.3 The Online Phase
5.4 The Multi-head Attention Model
6 The Experimental Setup
7 Comparison with Baselines
8 Experimental Results
9 Conclusion
References
Heterogeneous Device Resilient Indoor Localization Using Vision Transformer Neural Networks
1 Introduction
2 Related Works
3 Analysis of RSSI Fingerprints
4 Vision Transformers: Overview
5 VITAL Framework
5.1 Data Augmentation Module
5.2 Vision Transformer
6 Experimental Results
6.1 Indoor Paths and Smartphones
6.2 Hyperparameter Exploration
6.3 Comparison with State of the Art
6.4 Effectiveness of DAM
6.5 Results on Extended (New) Smartphones
7 Conclusion
References
Part IV Enabling Temporal Variation Resilience for ML-Based Indoor Localization and Navigation
Enabling Temporal Variation Resilience for ML-Based Indoor Localization
1 Introduction
2 Related Work
2.1 Types of Wi-Fi Localization
2.2 Countermeasures Against Age Deterioration of Localization Model
2.3 Signal Anomaly Detection
3 Wi-Fi Anomaly Detection with ``No-Sweat''
3.1 Key Idea of No-Sweat Detective System
3.2 Localization Protocol of No-Sweat Detective
3.2.1 Two Types of Observation
3.2.2 Challenge of No-Sweat Detective
3.2.3 System Overview
3.2.4 Access Point Anomaly Detector (APAD)
3.2.5 Reference Point Anomaly Detector (RPAD)
3.2.6 Transfer Learning Module (TLM)
3.3 Evaluation
3.3.1 Environmental Settings
3.3.2 Results and Discussions
3.4 Conclusion of This Section
4 Recalibration for Wi-Fi Fingerprinting Localization
4.1 Problem Setting
4.1.1 Formalization as Multi-class Classification
4.2 Group-Level Total Variation Regularization
4.2.1 Model Maintenance by Retraining
4.2.2 Designing Regularization Term
4.2.3 Cost of Learning
4.3 Experiments
4.3.1 Comparison Methods
4.3.2 Evaluation Metric
4.3.3 Shopping Mall Dataset (Uncontrolled Data)
4.3.4 Evaluation Results with Uncontrolled Data
4.4 Application: AP Movement Detection
4.5 Conclusion of This Section
5 Conclusion
References
A Few-Shot Contrastive Learning Framework for Long-Term Indoor Localization
1 Introduction
2 Background and Related Work
3 Siamese Network and Triplet Loss: Overview
4 STONE Framework
4.1 Overview
4.2 RSSI Fingerprint Preprocessing
4.3 Long-Term Fingerprint Augmentation
4.4 Convolutional Neural Encoder
4.5 Floorplan-Aware Triplet Selection Algorithm
5 Experiments
5.1 Experimental Setup
5.1.1 Fingerprinting Test Suite: UJI
5.1.2 Fingerprinting Test Suite: Office and Basement
5.1.3 Comparison with Prior Work
5.2 Experimental Results: UJI
5.3 Experimental Results: Office and Basement
5.4 Results: Sensitivity to Fingerprints per RP
6 Conclusion
References
A Manifold-Based Method for Long-Term Indoor Positioning Using WiFi Fingerprinting
1 Introduction
2 WiFi Fingerprinting for Indoor Positioning
3 Manifold-Based Nonlinear Regression for Indoor Positioning
3.1 Laplacian Eigenmap and Incremental Embedding
3.2 Common Issues and Remedies
3.2.1 Newly Added APs
3.2.2 Missing APs or RSS Measurements
3.3 Manifold-Based Indoor Positioning Framework
4 Experiments
4.1 Database Description
4.2 Results and Observation
4.2.1 Manifold Analysis
4.2.2 Floor Estimation
4.2.3 Indoor Position Estimation
4.2.4 Sensitivity Analysis
4.2.5 Computational Time
5 Conclusion
References
Part V Deploying Indoor Localization and Navigation Frameworks for Resource Constrained Devices
Exploring Model Compression for Deep Machine Learning-Based Indoor Localization
1 Introduction
2 Background and Related Work
3 CHISEL Framework
3.1 Data Preprocessing and Augmentation
3.2 Network Architecture
3.3 Model Compression
4 Experiments
4.1 Evaluation on UJIIndoorLoc Dataset
4.2 Evaluation on Compression-Aware Training
5 Conclusion
References
Resource-Aware Deep Learning for Wireless FingerprintingLocalization
1 Introduction
2 Toward Sustainable Green AI
2.1 Carbon Footprint of AI Models
3 Methodology for Calculating Model Complexity, Energy Consumption, and Carbon Footprint
3.1 Fully Connected Layer
3.2 Convolutional Layer
3.3 Pooling Layer
3.4 Total Number of FLOPs per Neural Network
3.5 Theoretical Computational Performance
3.6 Theoretical Energy Consumption
3.7 Calculating Theoretical Carbon Footprint
4 On Designing the PirnatEco Model for Localization
4.1 CTW2019 Dataset
4.2 DL Architecture Adaptation
5 Performance and Resource Consumption of DL Architectures for Localization
5.1 Evaluation Metrics
5.2 Energy Consumption and Carbon Footprint
6 Summary
References
Toward Real-Time Indoor Localization with Smartphones with Conditional Deep Learning
1 Introduction
2 Background and Related Work
2.1 Received Signal Strength Indicator (RSSI)
2.2 Indoor Localization Methodologies
2.3 Fingerprinting-Based Indoor Localization
2.4 Model Compression for Energy-Efficient Deployment
3 CNNLOC Framework Overview
3.1 Convolutional Neural Networks
3.2 Indoor Localization with CNNLOC
4 Localization Inference Analysis
5 Conditional Early Exit Models
6 QuickLoc Framework
6.1 QuickLoc CNN Model Design
6.2 QuickLoc CNN Model Training
6.3 Uncertainty Sampling Threshold
6.4 Post-deployment Configuration Adaptivity
7 Experimental Setup
7.1 Heterogenous Device Specifications
7.2 Indoor Paths for Localization Benchmarking
7.3 Comparison with Previous Work
7.4 Deployment and Evaluation
8 Experimental Results
8.1 Sensitivity Analysis for Uncertainty Sampling
8.2 Sensitivity Analysis on Device Heterogeneity
8.3 Analysis of Early Exit Path Configuration
8.4 Analysis of Inference Energy
8.5 Analysis on Memory Footprint
8.6 Analysis on Battery Life
8.7 Overall QuickLoc Performance
9 Generality of Proposed Approach
9.1 Depthwise Separable Convolutions-Based Network
9.2 Predicting Buildings and Floors for the UJIndoorLoc Dataset
10 Conclusions
References
Part VI Securing Indoor Localization and Navigation Frameworks
Enabling Security for Fingerprinting-Based Indoor Localization on Mobile Devices
1 Introduction
2 Background and Related Work
2.1 Received Signal Strength Indicator (RSSI)
2.2 Fingerprinting-Based Indoor Localization
2.3 Challenges with Indoor Localization
3 CNNLOC Framework Overview
3.1 Convolutional Neural Networks
3.2 Indoor Localization with CNNLOC
4 Localization Security Analysis
5 Problem Formulation
6 S-CNNLOC Framework
6.1 Offline Fingerprint Database Extrapolation
6.2 Inducing Malicious Behavior
7 Experiments
7.1 Experimental Setup
7.2 Experimental Results
7.2.1 Analysis of Noise Induction Aggressiveness
7.2.2 Comparison of Attack Vulnerability
7.2.3 Extended Analysis on Additional Benchmark Paths
7.2.4 Generality of Proposed Approach
7.2.5 Denoising Autoencoder-Based DNN Framework
7.2.6 Security-Aware DNN Training in the Offline Phase
8 Conclusions and Future Work
References
Index