Multipath-Assisted Multi-Occupant Vital Sign Sensing for Reflective In-Cabin Environments Using IR-UWB Radar Technology
- 주제(키워드) IR-UWB radar , multipath exploitation , vehicle cabin , occupancy detection , vital sign sensing , gradient boosting , angle-of-arrival , NLOS sensing
- 주제(DDC) 004.6
- 발행기관 아주대학교 일반대학원
- 지도교수 Young-Bae Ko
- 발행년도 2026
- 학위수여년월 2026. 8
- 학위명 석사
- 학과 및 전공 일반대학원 AI융합네트워크학과
- 실제URI http://www.dcollection.net/handler/ajou/000000036189
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
This thesis presents a geometry-guided multipath sensing framework for per-seat occupancy detection and breathing rate estimation in a four-seat vehicle cabin, using a sparse impulse-radio ultra-wideband （IR-UWB） radar deployment. Reliable in-cabin occupant monitoring is increasingly relevant to next-generation automotive safety, yet dense multipath propagation and overlapping echoes from passengers at similar ranges make per-seat signal attribution fundamentally difficult. Existing approaches suppress multipath components, discarding reflected signals that may carry the only recoverable physiological information for occluded occupants. The proposed framework exploits propagation geometry through virtual- transmitter path predictions to constrain candidate range bins on a per-seat basis, combined with a two-stage gradient-boosted classifier for within-row seat assignment. Stage 1 performs row-level occupancy detection by pooling candidate path features across both receivers, while Stage 2 resolves left, right, or both-seat assignment using inter-seat contrast features that cancel common-mode variation while preserving discriminative spatial signal. Breathing rate is subsequently estimated from the differential slow-time phase between receiver nodes, suppressing common-mode motion artifacts while preserving the respiratory component. The framework is evaluated on 117 recordings spanning all ten one- and two-seat occupancy configurations in a real vehicle cabin. The pipeline achieves a macro F1 of 0.878 for per-seat occupancy association, and estimates breathing rate at a mean absolute error of 2.35RPM overall. These results are achieved without synthetic reflectors, dense antenna arrays, or sensor counts beyond what is already present in production automotive UWB deployments. Security and privacy implications of passive physiological sensing are discussed, and design considerations for tamper-resilient and privacy-preserving deployment in safety-critical automotive systems are identified.
more목차
I. Introduction 1
1.1 Motivation 1
1.2 Problem Statement 2
1.3 Research Objectives 3
1.4 Contributions 3
II. Background and Literature Review 5
2.1 IR-UWB Radar Fundamentals 5
2.1.1 Radar Geometry: Monostatic, Bistatic, and Multistatic Configurations 5
2.1.2 Pulsed vs. Continuous Wave Radar and the UWB Advantage 6
2.1.3 The Channel Impulse Response 7
2.1.4 Sensing Sub-Millimeter Motion: Phase and Amplitude in IR-UWB . 8
2.1.5 Comparison with mmWave FMCW Radar 8
2.2 Multipath Propagation in Enclosed Spaces 9
2.2.1 The Nature of Multipath 9
2.2.2 The Vehicle Cabin as a Multipath-Dense Environment 9
2.2.3 Multipath as Interference: The Conventional View 10
2.2.4 Multipath as a Sensing Resource: The Emerging View 10
2.2.5 Virtual Transmitters and the Image Source Method 11
2.3 Related Work 12
2.3.1 Single-Occupant Vital Sign Sensing 12
2.3.2 Multi-Occupant and Vehicle Cabin Sensing 13
2.3.3 Multipath Exploitation for Localization and Sensing 14
2.3.4 Research Gap 14
III. System Model 16
3.1 Multistatic CIR Model 16
3.2 Virtual Transmitter and Passenger Geometry Model 16
IV. Methodology 19
4.1 System Overview 19
4.2 Feature Extraction 20
4.2.1 Signal Cleaning 20
4.2.2 Geometric Candidate Selection 21
4.2.3 Feature Extraction 24
4.3 Seat Classification 28
4.3.1 Classifier Selection and Training Configuration 29
4.3.2 Feature Aggregation: Row-Level Max-Pooling （Stage 1） 30
4.3.3 Stage 1 — Binary Row Detection 31
4.3.4 Feature Aggregation: Geometry-Weighted Pooling and Contrast （Stage 2） 31
4.3.5 Stage 2 — Three-Class Seat Assignment 33
4.4 Vital Sign Estimation 33
4.4.1 Phase Signal Model 34
4.4.2 Differential Phase and Common-Mode Rejection 34
4.4.3 Differential Phase Pre-processing 34
4.4.4 Spectral Analysis via FFT 35
4.4.5 CA-CFAR Peak Detection 36
4.4.6 Cross-Bin Voting and Candidate Selection 37
4.4.7 Ground-Truth Breathing Rate from ECG 38
V. Experimental Setup 40
5.1 Device Configuration and Deployment 40
5.2 Dataset 40
5.3 Evaluation Metrics 41
5.3.1 Occupancy Detection Metrics 42
5.3.2 Breathing Rate Estimation Metrics 42
VI. Results and Discussion 43
6.1 Occupancy Detection 43
6.1.1 Overall Performance 43
6.2 Receiver Contribution Analysis 44
6.3 Feature Importance Analysis 45
6.4 Ablation Study 47
6.5 Breathing Rate Estimation 49
6.6 Discussion 50
6.6.1 Comparison with Neural Network Baselines 50
6.6.2 Comparison with Prior Systems 54
VII.Conclusion 57
7.1 Summary 57
7.2 Limitations 57
7.3 Future Work 58
References 60
