Gait analysis is a critical biomarker for detecting locomotion abnormalities, identifying unique gait patterns, and implementing and evaluating rehabilitation interventions. Existing gait analysis systems provide high accuracy and resolution but lack portability and accessibility, are limited to indoor settings, and are expensive. These limitations are addressed through wearable gait measurement systems based on custom-built inertial measurement units (IMUs) that deploy multiple sensors around the body to track multiple parts of the body (e.g., neck, lower back, and feet). However, they may be less preferable in real-world applications because of human factors (e.g., convenience, appearance, comfort, or fear of social stigma). In this thesis, we employ an alternative and less obtrusive approach. We developed a software application for a standard smartphone that records and analyzes movement data from its embedded IMU sensors. The main objective is to design an automated smartphone-based gait monitoring system to monitor and characterize gait of various subjects, including people suffering from neurological movement disorder and healthy controls. Human gait is a complex and highly variable mechanism that greatly differs between individuals. We investigate whether these variations can be objectively measured and characterized using the new system during three key functional clinical assessment tests: chair sit-to-stand test (CST), timed-up-and-go (TUG) and 10-meter walking tests. These tests are used to measure endurance, predict falls, measure balance, and predict cognitive decline in older adults. The proposed smartphone-based gait analysis system has two novelties: the first novelty is an adaptive threshold peak detector to detect key events in CST, TUG, and 10-meter walk data. The second novelty is context awareness, that is, knowing who is performing the functional test? What test is being performed? In what settings is the test conducted indoors or outdoors? Does the filtering frequency for the given data based on environment activity be determined? We tested and validated our adaptive peak detector in young, old, stroke and PD subjects, with an accuracy greater than 97%. The results obtained from all functional tests are validated against camera-recorded video where the mean absolute error (MAE) is < 50 ms. Our system is extensively tested and validated with data recorded from lower back and its performance is expected to vary if phone is positioned in the pocket or any other body location. To begin to test this, we investigated sampling frequency, window size, sensor type within the context of surface classification and person identification. Overall, our system is up to par with the existing state of the art systems and offers portable, convenient, inexpensive and accessible solution. It is a step forward towards personalised and continuous automatic monitoring and characterisation of CST, TUG and 10-meter tests in free living environments.
Automating gait analysis using a smartphone
Sher, A. (Awdur). 2023
Traethawd ymchwil myfyriwr: Traethawd Ymchwil Doethurol › Doethur mewn Athroniaeth