Page 755 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Industry, Innovation, and Infrastructure: The use case leverages advanced AI and scalable
systems to modernize care delivery. By utilizing existing CCTV infrastructure, it introduces
innovative solutions without the need for costly hardware, making implementation both
practical and efficient. 4.9: Accessibility

Reduced Inequalities: By offering low-cost, AI-powered care tools in under-resourced care
facilities, the solution helps bridge service gaps and promotes equity. It ensures that individuals
in underserved areas benefit from the same quality of care and digital support as those in more
developed settings, reducing disparities and fostering inclusion [4].

2�3 Future Work

• Model enhancement: Integrate audio and biosensor (smart watch/band, Internet of
Things (IoT) devices) data for multimodal behavior analysis.
• New applications: Expand to elderly care, childcare, and school safety.
• Continual learning: Improve accuracy by retraining with new data.
• Deployment at scale: Extend rollout through national and global public-private
partnerships.
• Collaborations: Align with International Telecommunication Union (ITU) efforts on AI
standards for healthcare and safety monitoring.[5]
• Sharing data: Currently, due to the sensitive nature of the data (involving individuals
with developmental disabilities), our datasets are private and used under strict ethical
supervision.

3 Use Case Requirements

REQ-01: It is critical that the system deploys a lightweight AI model optimized for edge
computing, capable of operating on devices with a maximum computational capacity of 100
TOPS (Tera Operations Per Second) or lower.

REQ-02: It is critical that the system performs real-time video analysis with behavior detection
accuracy > 85% to ensure reliable monitoring and intervention capabilities..

REQ-03: Support for privacy-preserving behavior monitoring (meta info only).
• CareVia is designed with privacy by design principles. The system does not store or
transmit raw video externally.
• Anonymization is applied at the edge level via facial blurring and meta-information
extraction.
• All processing stays on-device (within the AI Box) unless users explicitly export anonymized
reports.
• The platform complies with local privacy laws and follows European Union (EU)-inspired
data minimization principles.
• The system only shares timestamped behavior types and non-identifiable metadata across
the platform.
• This ensures privacy is protected, even in environments where closed-circuit television
(CCTV) use is highly regulated.
REQ-04: It is expected that the system seamlessly integrates with existing CCTV infrastructure
to minimize deployment costs and leverage legacy hardware.

719