With an aging population, the number of patients with gait disorders is steadily increasing. Since the diseases causing gait disturbances span multiple clinical departments and effective screening tests are lacking, timely diagnosis remains a significant challenge. In this study, we propose a novel method for estimating lumbar spinal canal stenosis (LCS)–one of the most prevalent causes of gait disorders–from patient gait video sequences. Conventional video-based gait analysis primarily utilizes skeletal (wireframe) models or silhouette features. However, skeletal models fail to capture detailed information such as spinal curvature or joint torsion, while silhouette-based approaches are prone to incorporating irrelevant information like clothing. To address these limitations, our method first estimates a 3D skin mesh model from the gait video and then fits an anatomical skeleton model to it. This process enables the extraction of fine-grained postural features, including 3D joint coordinates and angles that are difficult to obtain with conventional methods. We collected gait videos from patients with and without gait disorders for evaluation. The results demonstrate that our approach achieves significantly higher estimation accuracy compared to conventional methods based on wireframe or silhouette features.