AnomGait: Data-Driven Extraction of Movement Features Using Contrastive Learning
摘要
To describe movement patterns in gait, interpretable handcrafted features such as step length and step width are typically used. Current movement analysis tools allow the generation of rich, high-dimensional data for each individual, which is then, however, again reduced to traditional, single-dimension parameters. We here apply contrastive learning to high-dimensional gait data sets from neurological patients and healthy individuals and attempt to distill data-driven features to improve separation.