bachelorThesis
Generación de la biomecánica del movimiento de extremidades inferiores
Fecha
2021-04-23Autor
Benenaula Armijos, Stalin Javier
Trelles Peralta, Milton Damian
Institución
Resumen
Human beings, in their quest to improve the quality of life in people who suffer from gait disturbances, have
developed technologies capable of identifying patterns and characteristics of different conditions that deteriorate
mobility. Although, some studies explore invasive methods such as electromyography, the use of sensors and/or
markers for the analysis and evaluation of pathological gaits, there is still little research that addresses methods
that do not invade the body, given that in current times it is an essential approach.
The purpose of this study is to develop a non-invasive system, based on vision techniques and artificial
intelligence capable of generating spatio-temporal parameters of the biomechanics of movement of the lower
extremities from normal or pathological gaits such as hemiparetic and paraparetic, as well as the analysis and
classification of these gaits.
The methodology used consists of capturing RGB images in people who perform several cycles of the
normal, hemiparetic and paraparetic gaits. These images are processed by using models like OpenPose and
PoseNet to estimate the pose. Then, cutting, synchronization, filtering, normalization and 2D analysis techniques
are applied, as well as new approaches such as Skeleton Gait Energy Image (SGEI) to characterize the gait.
Finally, through algorithms such as Convolutional Neural Network (CNN) or Support Vector Machine (SVM),
the system is trained to classify the analyzed gaits.
As a result, it is possible to generate the parameters of stride length, cadence, stride width, step time, gait
speed, front body posture inclination and angles of the lower extremities of the human body of the 3 gaits using
a non-invasive system approach, additionally experimental results show high efficiency in the classification of
the gaits with a 98.57 % using OpenPose and a 98.15 % with PoseNet.