File
M. Ota1, H. Tateuchi1, T. Hashiguchi2, N. Ichihashi3
1Kyoto University, Department of Preventive Physical Therapy, Human Health Sciences, Graduate School of Medicine, Kyoto, Japan, 2Mixi Incorporated, Tokyo, Japan, 3Kyoto University, Department of Physical Therapy, Human Health Sciences, Graduate School of Medicine, Kyoto, Japan
Background: The human tracking algorithm called OpenPose is an open software that can detect joint positions for free except for commercial use. OpenPose is valid and reliable when used to measure squat kinematic. Within device test-retest reliability was almost complete, and OpenPose values were associated with VICON values significantly. However, the validity of gait analysis using OpenPose has not been clarified yet.
Purpose: The purpose of this study was to verify the validity of gait analysis using OpenPose.
Methods: Twenty-four healthy subjects participated in this study (17 males and 7 females; age: 26.0 ± 4.7 years; height: 167.6 ± 7.7 cm; mass: 61.1 ± 10.8 kg). The Ethics Committee of Kyoto University approved this study (#R1823). The motion task were treadmill walking and running. Each velocity is defined as follows: 2.5, 4.0, 5.5, and 8.5 km/h. Kinematic measurements by three-dimensional motion analysis systems were recorded using VICON at a sampling rate of 60 Hz and a low-pass filter with a 6 Hz cut off. Simultaneously, the images were taken with a digital camera from the right and back side at same sampling rate and low-pass filter as VICON. After the images were processed with OpenPose, segment and joint angles were measured from estimated joint points. The angles on the frontal plane for pelvis and hip joint on right side, and the angles on the sagittal plane for hip, knee, and ankle joints on right side were measured. Peak angles were calculated using OpenPose and VICON, and the ranges during one gait cycle were calculated. The mean values of four successful trials were used for analyses. To confirm the validity, the linear regression analysis and intra-class correlation coefficients [ICC (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. The level of significance was set at 0.05.
Results: Regarding the hip flexion-extension ranges, and most parameters of knee and ankle, associations and ICCs (2, 1) were significant, and there was not proportional bias. In particular, the associations and ICCs (2, 1) for the ranges of those joints were moderate or higher. However, regarding many parameters of hip except for the flexion-extension ranges, and most parameters of pelvis, associations and ICCs (2, 1) were nonsignificant or insufficient, and there were fixed and proportional biases.
Conclusion(s): Validity of gait analysis on the frontal plane using OpenPose was insufficient. However, most parameters on the sagittal plane during walking and running except for the hip flexion and extension angles obtained by OpenPose and VICON were good associations and agreements. In other words, gait analysis on the sagittal plane using OpenPose showed enough validity. It was suggested that it should be possible to measure the lower limb angles on the sagittal plane using OpenPose from images captured with one digital camera.
Implications: OpenPose is a valid, low cost and easy to operate tool that can facilitate clinician access to kinematic data.
Funding, acknowledgements: This study was supported by a joint research fund from Smart Health Incorporated.
Keywords: Gait analysis, Human pose tracking algorithm, OpenPose
Topic: Innovative technology: robotics
Did this work require ethics approval? Yes
Institution: Kyoto University
Committee: The Ethics Committee of Kyoto University
Ethics number: R1823
All authors, affiliations and abstracts have been published as submitted.
