Deep vision system for clinical gait analysis in and out of hospital

Abstract

To follow-up Parkinson’s disease (PD) progress, clinical gait analysis is performed with the precise measuring equipments (e.g. IMU, electric walkway, etc.). However, the existing gait analysis methods have a limitation such that patients must visit a certain space in hospital for the checkup. For clinical gait analysis in and out of hospital, we propose a baseline model of ‘deepvision’ system, which can estimate 15 clinical gait parameters measured from electric walkway named GAITRite. We constructed 3D convolution layers which have skip connections to grasp spatio-temporal characteristics of the walking behavior with an effective manner. Afterwards, we validated the method with scripted walking videos, and achieved the following results: error range of temporal and spatial parameters as 32–71 ms, 1.6–6.7 cm respectively, and error for cadence, velocity and functional ambulation profile as 7.0 steps/min, 4.1 cm/min, and 4.9 points respectively.