Por favor, use este identificador para citar o enlazar este ítem:
http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/2582| Real-time Performance Evaluation of Yoga Poses Using the NVIDIA Jetson Nano: A Comparative Study Involving Stereo Vision and Body Angles Estimation Methods | |
| Eric Williams-Linera | |
| JUAN MANUEL RAMIREZ CORTES | |
| Acceso Abierto | |
| Atribución-NoComercial-SinDerivadas | |
| Yoga Pose evaluation MediaPipe Kinect Stereo Vision Jetson Nano Body Angles | |
| Practice of Yoga has seen increased popularity in recent years as it provides several health benefits through physical, mental, and spiritual practices. While several online resources are available for people to perform yoga at home without requiring an instructor, unsupervised training can increase risk of injury, as users are not provided suggestions on how to improve. This thesis project proposes a system, hosted on an NVIDIA Jetson Nano, to evaluate user performance of yoga poses in real time. The MediaPipe Pose framework was enabled to perform pose detection to estimate the 3D location of several joints of the human body, which subsequently allowed to perform pose evaluation. Pose evaluation was performed by estimating several joint angles of a particular user and comparing them with a reference pose. The poses applied to this work were Goddess, Warrior II and Tree, which can be considered easy to perform and have been applied to other works in the literature. This work involved the implementation of both single camera and stereo vision systems, using one and two cameras, respectively, to evaluate their respective performances for human pose estimation and evaluation. Similarly, two methods to estimate body joint angles were implemented, the vector dot product and a procedure based on Inverse Kinematics. The estimated joint angles allowed to design and implement a protocol to assess users in real time while performing yoga. Through a Graphical User Interface, the proposed system provides users with a score ranging from 0 to 100, which is based on the percentage error between user angles and the reference pose. Additionally, the system enables a color scale system that visually indicates practitioners how to improve as they perform a pose. Pose detection and angle estimation results of the implemented frameworks were validated with the aid of a Kinect V1 device. Results prove that Stereo Vision outperforms the single camera system in terms of accuracy for 3D pose detection and angle estimation and is, therefore, more reliable for providing correct feedback. The proposed system will be of aid to individuals that practice yoga as it will minimize injury risk while improving physical and mental health. | |
| Instituto Nacional de Astrofísica, Óptica y Electrónica | |
| 2024-09 | |
| Tesis de maestría | |
| Inglés | |
| Estudiantes Investigadores Público en general | |
| Williams Linera, E., (2024), Real-time Performance Evaluation of Yoga Poses Using the NVIDIA Jetson Nano: A Comparative Study Involving Stereo Vision and Body Angles Estimation Methods, Tesis de Maestría, Instituto Nacional de Astrofísica, Óptica y Electrónica. | |
| ELECTRÓNICA | |
| Versión aceptada | |
| acceptedVersion - Versión aceptada | |
| Aparece en las colecciones: | Maestría en Electrónica |
Cargar archivos:
| Fichero | Tamaño | Formato | |
|---|---|---|---|
| WILLIAMSLE_ME.pdf | 11.37 MB | Adobe PDF | Visualizar/Abrir |