This research introduces SKI Models that combine skeleton data with vision-language models to better understand Activities of Daily Living (ADL) videos.

• Addresses key limitations of existing vision models when analyzing subtle human movements
• Leverages 3D skeleton information to distinguish similar-appearing activities
• Improves performance across multiple viewpoints common in healthcare monitoring
• Creates embeddings that better capture the nuances of daily activities

This advancement has significant implications for healthcare monitoring systems, enabling more accurate assessment of patient functionality, independence, and rehabilitation progress through automated observation of daily activities.

SKI Models: Skeleton Induced Vision-Language Embeddings for Understanding Activities of Daily Living