Verification of validity of gait analysis systems during treadmill walking and running using human pose tracking algorithm.

Affiliation

Ota M(1), Tateuchi H(2), Hashiguchi T(3), Ichihashi N(4).
Author information:
(1)Department of Preventive Physical Therapy, Human Health Sciences, Graduate School of Medicine, Kyoto University, Kyoto, Japan. Electronic address: [Email]
(2)Department of Preventive Physical Therapy, Human Health Sciences, Graduate School of Medicine, Kyoto University, Kyoto, Japan.
(3)Mixi Incorporated, Tokyo, Japan.
(4)Department of Physical Therapy, Human Health Sciences, Graduate School of Medicine, Kyoto University, Kyoto, Japan.

Abstract

BACKGROUND: The human tracking algorithm called OpenPose can detect joint points and measure segment and joint angles. However, the validity of gait analysis using OpenPose has not been examined yet. RESEARCH QUESTION: What is the validity of OpenPose-based gait analysis? METHODS: Twenty-four healthy young people participated in this study. The participants were assessed during walking and running. Pelvic segment angles, and hip, knee, and ankle joint angles during treadmill walking and running were measured using VICON. Simultaneously, images were captured using digital cameras from the right and back sides. After processing with OpenPose, the corresponding angles were measured from the estimated joint points. To validate these estimations, linear regression analysis was performed, and intraclass correlation coefficients [ICCs (2, 1)] between the data obtained by OpenPose and VICON were calculated. Furthermore, the agreement between the data obtained by OpenPose and VICON was assessed by Bland-Altman analysis. RESULTS: For most ranges of motion (ROM) in the sagittal plane, the hip, knee, and ankle joints had large coefficients of determination, without proportional biases. For most peak angles in the sagittal plane, the knee and ankle joints had large coefficients of determination without proportional biases, although the hip joint had nonsignificant coefficients of determination and proportional biases. In particular, for the hip flexion-extension ROM and peak knee flexion angle during running and the knee ROM during slow walking, the ICCs showed good to excellent agreement. However, for the parameters of the pelvis and hip joint in the frontal plane, there were nonsignificant coefficients of determination and poor ICCs with fixed and proportional biases. SIGNIFICANCE: The lower limb ROM in the sagittal plane during gait can be measured by the OpenPose-based motion analysis system. The markerless systems have the advantage of being more economical and convenient than conventional methods.