깊이 데이터 인코딩과 동적 스케일 보정을 적용한 다중 카메라 기반 3차원 보행 분석 시스템

Abstract

This study proposes a three-dimensional (3D) gait analysis system designed to overcome the spatial limitations and distance-dependent accuracy degradation of conventional systems by employing an orthogonal arrangement of multiple depth cameras. To improve data efficiency, the system encodes depth information in RGB format and applies k-means clustering to reduce noise. Two-dimensional joint coordinates, extracted using a YOLOv8-pose model, are scale-corrected according to the subject's height and then fused with the corresponding Z-values from the registered depth images to reconstruct 3D coordinates. The proposed approach is significant in that it integrates low-cost hardware with efficient algorithms to provide accurate and convenient gait analysis, particularly in constrained clinical environments.