Machine Learning Aided Millimeter Wave System for Real Time Gait Analysis
Date of Award
2022
Degree Name
Ph.D. in Engineering Management, Systems, and Technology
Department
Department of Engineering Management, Systems, and Technology
Advisor/Chair
Vamsy Chodavarapu
Abstract
Gait analysis measures the walking biomechanics and identifies the abnormality in regular walking patterns. This information is useful for clinical and rehabilitation purposes. The walking patterns can be observed using wearables, cameras, radars and Light Detection and Ranging (LiDAR). The LiDAR and cameras are expensive. Furthermore, cameras invade the privacy of a user. Wearables are beneficial in taking outdoor gait readings. But, they are cumbersome to wear for longer durations and have limited accuracy. The Millimeter Wave (MMW) radars have attracted significant attention in gait analysis because of their affordability, portability, simplicity, privacy and ability to operate in various ambient climate conditions. This work uses a low-cost MMW radar to develop a portable fall detection system using gait analysis. It examines the performance of popular Machine Learning (ML) techniques for gait analysis, including Support Vector Machine (SVM), Decision Tree (DT) and Neural Network (NN) for fall detection. The results indicate that NN achieves 99.79% training accuracy compared to 98.85% training accuracy for DT and 98.27% accuracy for SVM. The same trends are followed in testing accuracy. Therefore, the proposed fall detection system consists of MMW radar, NN-based Long Short-Term Memory (LSTM) and a low-cost NVIDIA Jetson nano-board, which shows promising results in terms of fall detection. We propose a novel solution, MMW radar system, for Human Activity Recognition (HAR). The mmGait combines micro-Doppler signatures of different activities and the skeleton pose estimation for 19 different joints. The proposed system uses a low-cost MMW radar, Kinect V2 sensor and Convolutional Neural Network (CNN) to classify five different activities. It can identify single or multiple activities in different environments. Furthermore, it can classify activities for different subjects in the same environment. The experimental results show that proposed system can classify different gait activities with an accuracy of 98.8%.
Keywords
Electrical Engineering, mmWave Radar, Gait Analysis, Machine Learning, Human Pose Detection, Nvidia Jetson Nano
Rights Statement
Copyright © 2022, author.
Recommended Citation
Alanazi, Mubarak Alayyat, "Machine Learning Aided Millimeter Wave System for Real Time Gait Analysis" (2022). Graduate Theses and Dissertations. 7174.
https://ecommons.udayton.edu/graduate_theses/7174
