An Accuracy Enhancement Method for Markerless Gait Analysis Using Computer Vision

Gait analysis has been applied in a wide range of fields, including medicine and rehabilitation, security, and sports science. Conventional gait measurement techniques have relied on wearable devices and 3D motion capture systems. However, these methods face challenges such as the need for specialized facilities and expensive equipment, long setup and post‐processing times, and the possibility of the devices themselves affecting natural gait. To address these issues, markerless gait analysis using computer vision (CV) has recently gained attention as a low‐invasive and cost‐effective alternative. With the advancement of CV technologies, 2D pose estimation methods such as OpenPose have been released, enabling the extraction of human joint coordinates from video footage alone. Nevertheless, in real‐world environments, detection errors often occur, leading to significant inaccuracies in gait analysis. In this paper, we propose a two‐stage filtering method that combines second‐order difference Kalman filtering for dynamic estimation and a Butterworth filter for residual noise reduction to correct the coordinate detection errors caused by OpenPose. Furthermore, the effectiveness of the proposed method is demonstrated through a comparative analysis with a conventional optical marker‐based motion capture system.