검색 상세

Human Locomotion Recognition, STS Transition Phase Estimation and Gait Phase Recognition of Level-ground Walking Using Wearable IMU Sensors

웨어러블 IMU 센서를 이용한 수평 지면 보행의 인간 운동 분류, STS 전환 단계 추정 및 보행 단계 인식

초록 (요약문)

In this thesis, we present a novel approach to estimating the intent of human movement during a user performing different locomotion modes, such as sitting, standing, sit-to-stand (SiSt), stand-to-sit (StSi) transitions, and gait phases of level ground walking using wearable inertial measurement units (IMUs). Our experiment involved 18 healthy subjects with IMUs attached to their chest, thigh, and ankle to collect comprehensive sensor data of subjects performing those movements. The data processing was meticulously carried out, focusing on maximizing identifying transition phases. Our approach utilized a Multi-Layer Perceptron (MLP) neural network algorithm in a supervised learning framework. The MLP was trained with data from 14 subjects and validated with the data from the four subjects. The results from offline and real-time predictions were promising, showing that MLP can effectively predict five locomotion modes such as sit, stand, SiSt, StSi, and walking; 11 transition phases in SiSt/StSi movements, and seven distinct gait phases in level-ground walking.

more

초록 (요약문)

이 논문에서는 사용자가 앉기, 서기, 앉았다 일어서기(SiSt), 일어서서 앉기(StSi) 전환과 같은 다양한 운동 모드를 수행하는 동안 인간 움직임의 의도를 추정하는 새로운 접근 방식을 제시합니다. 웨어러블 관성 측정 장치(IMU)를 사용하여 평지 걷기의 보행 단계를 측정합니다. 우리의 실험에는 가슴, 허벅지, 발목에 IMU를 부착한 건강한 피험자 18명이 참여하여 이러한 움직임을 수행하는 피험자의 포괄적인 센서 데이터를 수집했습니다. 데이터 처리는 전환 단계 식별을 극대화하는 데 중점을 두고 꼼꼼하게 수행되었습니다. 우리의 접근 방식은 지도 학습 프레임워크에서 MLP(Multi-Layer Perceptron) 신경망 알고리즘을 활용했습니다. MLP는 14개 주제의 데이터로 훈련되었으며 4개 주제의 데이터로 검증되었습니다. 오프라인 및 실시간 예측 결과는 유망했으며 MLP가 앉기, 서기, SiSt, StSi 및 걷기와 같은 5가지 운동 모드, SiSt/StSi 움직임의 11가지 전환 단계와 평지 걷기의 7가지 개별 보행 단계를 효과적으로 예측할 수 있음을 보여주었습니다.

more

목차

I Introduction 1
A Research Background 1
B Related Works 4
C Research Purpose 5
D Thesis Outline 6
II Methodology 8
A Experiment Protocol and Data Collection 8
a Participants Information 8
b IMU Placement and Orientations 9
c Data Acquisition Protocols 13
B Data Preprocessing 15
a Start/end Point Detection Algorithm for SiSt/StSi Transitions 16
b Trimming and Labeling of Transition Phases 18
c Gait Event Detection Algorithm for Level Ground Walking 22
d Walking Data Trimming and Labeling of Gait Phases 25
C Input Feature Selection and Dataset Preparation 27
a Input Features for Locomotion mode Classification 28
b Input Features for SiSt/StSi Transition Phase Estimation 29
c Input Features for Gait Phase Recognition of Level-ground Walking 31
III Model Parameters and Training 33
A Proposed Neural Network Algorithm 33
a Locomotion Mode Classification 33
b SiSt/StSi Transition Phase Estimation 33
c Gait Phase Recognition of Level-ground Walking 34
IV Results and Discussion 36
A Locomotion Mode Classification Model Results 36
B SiSt and StSi Transition Phase Estimation Model Results 37
C Results of the Gait Phase Recognition of Level-ground Walking 39
V Conclusion and Future Work 50
References 62

more