The paper from Design Lab “Knee Flexion/Extension Angle Measurement for Gait Analysis using Machine Learning Solution ‘MediaPipe Pose’ and its Comparison with Kinovea®” was presented at University Scholar Conference : Engineering and Innovation organized by Association of Mechanical Engineering Students (AMES) at Kathmandu University.
Abstract
Human Gait Analysis is crucial in diagnosis, monitoring, treatment as well as rehabilitation. It requires measurement of gait parameters like joint angles, angular motion, ground reaction force (GRF), momentum, step width, step length, velocity, etc. This study introduces a marker-less, easy-to-use approach for measuring one of the gait parameters i.e., knee flexion/extension angle, and verifies it with an existing standard marker-based approach. Knee flexion/extension angle is calculated via Machine Learning (ML) pose estimation model “MediaPipe Pose” and Computer vision (CV) without the use of markers. For the verification of accuracy, the obtained values are compared with the data on each video frame obtained from Kinovea® which is a marker-based motion analysis software. High correlations (r = 0.941) between Kinovea® and MediaPipe Pose were observed. Similarly, the mean absolute error of knee angles between MediaPipe Pose and Kinovea® was 5.88 degrees. The research shows that knee flexion/extension angle can be successfully and accurately measured using ML Solution for high-fidelity body pose tracking. Professionals involved in the field of biomechanics, sports medicine, physiotherapy as well as other medical fields can use this study as an alternative marker-less approach for knee flexion/extension angle measurement for gait analysis.
The paper is from the funded Research at Design Lab “Machine Learning Application on Development of Human Gait Analysis System for Clinical Purpose, using data from Computer Vision and Embedded Sensors“