Date of Award
Fall 11-24-2025
Document Type
Thesis
Publication Status
Version of Record
Submission Date
December 2025
Department
Computer and Electrical Engineering and Computer Science
Degree Name
Master of Science (MS)
Thesis/Dissertation Advisor [Chair]
Behnaz Ghoraani
Abstract
This thesis presents an advanced deep learning framework for markerless 3D gait assessment using monocular video, aimed at the early detection of Alzheimer’s disease (AD). Recognizing gait abnormalities as potential preclinical biomarkers, the study leverages video-based human pose estimation to derive detailed spatiotemporal gait parameters without the need for specialized laboratory equipment. The proposed system adapts the MeTRAbs (Metric-Scale Truncation-Robust Heatmaps for Absolute 3D Human Pose Estimation) architecture to reconstruct metric-scale 3D human joint coordinates from RGB video recordings. A complete processing pipeline is developed to extract stride length, cadence, stance and swing phases, and gait variability, validated against the Zeno™Walkway an established gold-standard, pressure-sensitive gait analysis system.
Experiments conducted on custom datasets, recorded using monocular cameras under varied indoor conditions, demonstrate strong correlations for temporal gait parameters and moderate agreement for spatial measures. Quantitative evaluations yield mean absolute errors within practical thresholds for cadence (0.96 steps/min) and step time (0.009 s), while correlation analyses confirm high reliability (r > 0.9) for temporal features. These results affirm that monocular video-based gait analysis can achieve clinically meaningful accuracy using low-cost, accessible technology.
The findings validate the feasibility of employing advanced neural architectures for vision-based gait analysis as a non-invasive, scalable tool for early AD screening. The study contributes to bridging the gap between laboratory precision and real-world accessibility, highlighting the potential of computer vision in advancing digital biomarkers for neurodegenerative disease assessment.
Recommended Citation
Kushwaha, Abhishek Singh, "ADVANCED DEEP NEURAL NETWORK ARCHITECTURES FOR 3D GAIT ASSESSMENT FROM VIDEO IN THE EARLY DETECTION OF ALZHEIMERS DISEASE" (2025). Electronic Theses and Dissertations. 199.
https://digitalcommons.fau.edu/etd_general/199