GaitMoText: A Multimodal Dataset for Clinical Gait Analysis compromising Optical Motion Capture and Textual Gait Analyses from Patients undergoing Total Knee Arthroplasty

GaitMoText is a multimodal dataset designed to advance automated clinical gait analysis by combining high-quality motion capture data with expert textual gait assessments.

The dataset contains recordings from 23 patients undergoing total knee arthroplasty (TKA), collected both pre-operatively and six weeks post-operatively during a standardized 6-minute walking test (6MWT). Motion data was captured using the SIMI Motion markerless system and is provided in multiple representations, including 3D keypoint trajectories, joint angle time series, and SMPL-based motion parameters.

In addition to quantitative motion data, the dataset includes rich qualitative annotations in the form of clinical gait assessments. Each recording session is annotated by multiple expert physiotherapists, resulting in detailed textual descriptions of gait abnormalities and compensatory mechanisms. The annotations are available in both German and English.

This dataset enables research at the intersection of biomechanics, computer vision, and natural language processing, supporting tasks such as automatic gait assessment, anomaly detection, rehabilitation monitoring, and multimodal motion understanding.

The dataset is fully anonymized and contains only privacy-preserving skeletal and parametric representations. Raw video data is not included.

For more details, please refer to the associated publication.