BSKI: Advancing Vision-Language Models for Understanding Activities of Daily Living – Accepted at AAAI 2025

We are excited to share that our paper, “BSKI Models: Skeleton-Induced Vision-Language Embeddings for Understanding Activities of Daily Living,” has been accepted at AAAI 2025. This research introduces BSKI, a Skeleton-Induced Vision-Language model designed for recognizing Activities of Daily Living (ADLs), with critical applications in healthcare, elderly care, and behavioral monitoring.

Enhancing Health Monitoring with BSKI

Monitoring ADLs is essential for early detection of health anomalies, fall risks, and behavioral changes that may indicate underlying medical conditions. While existing vision-based models analyze daily activities, they often lack robustness in recognizing fine-grained actions and biomechanical cues. BSKI addresses these challenges by leveraging 3D skeleton-based representations alongside vision-language embeddings to enhance ADL understanding.

Key Contributions of BSKI:

• Skeleton-Induced Vision-Language Modeling: Integrates skeletal pose estimation with vision-language embeddings to improve activity recognition.
• Enhanced ADL Detection for Senior Care: Facilitates early intervention in fall risks and behavioral changes, promoting independent living.
• Real-Time Insights for Caregivers & Healthcare Professionals: Provides continuous monitoring to ensure timely medical assistance.

Impact and Applications

BSKI has the potential to revolutionize elderly care by enabling continuous, real-time ADL tracking. This can enhance personalized healthcare, improve patient safety, and support independent living while assisting caregivers in providing proactive care.

Acknowledgments

This work was developed by Arkaprava Sinha, Dominick Reilly, Francois Bremond, Pu Wang, and Srijan Das, and will be presented at AAAI 2025.

📄 Read more about BSKI: Preprint

Stay tuned for further updates on AI-driven advancements in healthcare and ADL monitoring.