Abstract
Maintaining a high standard of life, especially for the elderly, enables them to live their lives withminimum fear of sudden accidents. Falling is one of the most important problems that affect elderly
lives. Falling causes injuries that may be fatal or decrease the functional ability and the quality of life.
Predicting and preventing falls before they happen makes a critical difference, enabling optimal self-care
for elderly people. While the main causes of falling are related to postural sway and walking, determining
abnormalities in one or both of these activities can be informative in predicting the likelihood of having
a fall. A need exists for a gait and postural sway analysis system that is easy to use, readily available,
and inexpensive
This PhD thesis investigates using vision data, as an inexpensive option, to estimate gait and sway
movements from video recordings. It also investigates measuring human gait and postural sway parameters
from the estimated movements within acceptable accuracy when compared to the gold standard
measurements, Vicon and force plate. Furthermore, based on analysing the measured gait and sway
parameters for a different age groups, an increasing risk of having a fall is predicted.
Defining the changes in the pattern of gait and sway that are considered as signs for deterioration
in an individual’s health status is important for predicting the risk of a fall. Monitoring these changes,
which are continuously occurring and are inevitable while a person is aging, allows early intervention to
avoid possible fall accidents well before they happen. Seeking help and including proper exercises in the
daily life are examples of appropriate intervention to avoid possible fall accidents.
First, a dataset of gait and sway activities is devised and recorded for two groups of people, elderly
people over fifty years and younger healthy athletes. The dataset part for the healthy athletes group of
people is considered as the ground truth part. Vision data is the main data type in the two parts. The
ground truth part of the dataset contains centre of pressure displacements from a force plate as well as
the joint movements in the three dimensional (3D) space from a Vicon motion capture system.
Second, a computational learning-based model to estimate the body’s postural sway from vision data
is proposed and validated. Another model to estimate human gait from vision data is also built and
validated using the ground truth part of the dataset. The estimated gait and sway movements are used to
measure defined parameters that are used to analyse the gait and sway. The two computational models
achieved high correlation between the estimated gait and sway signals from video recordings compared
to their corresponding signals extracted from the force plate and the Vicon system.
Third, the gait and sway proposed models are used to estimate gait and sway movements for the
group of elderly people, then the gait and the sway parameters are measured. Based on the measured
sway parameters, each elderly subject is assigned to a sway-age group. Also, for the measured gait
parameters, each elderly subject is assigned to a gait-age group. The chronological age of the subjects is
compared with the estimated age from sway and gait movements toward analysing the fall risk.
From the presented work, vision data can be used to estimate humans’ gait and postural sway with
an acceptable accuracy that has reached 90% for the estimated postural sway and gait referenced to
the measurements computed from the ground truth data. The proposed methods are then used on the
elderly people data to estimate their sway and gait movements. Using these estimations, sway and
gait parameters are measured. From these parameters, each elderly subject is clustered into gait/sway
age group. Comparing subject’s gait/sway age group with his/her chronological age group defines a
likelihood risk of having a fall that gives the opportunity to intervene to improve maintaining the balance
to avoid fall accidents.
| Date of Award | 2020 |
|---|---|
| Original language | English |
| Supervisor | Roland Goecke (Supervisor) & Gordon WADDINGTON (Supervisor) |