Proposing Human-Centered Monitoring Framework Characterizing Contexts with Vision-Based Edge AI

Amidst global aging in Japan, there is a significant shift from traditional facility-based care to home-based care due to shortages of care facilities and personnel. While home living is often preferred, this places a substantial burden on family caregivers in providing daily care and supervision for the elderly. In prior studies, methods such as elderly “mind” sensing through voice-based dialogue systems and quality assessment techniques were proposed for the elderly’s in-home activities using skeletal sensing technology. However, the recognition of changes in the elderly such as facial expressions, body posture, and behaviors is essential for monitoring elderly individuals at home and has not yet been fully developed. Hence, we proposed non-verbal features in the context of human-centered recognition. Multiple pre-trained models were integrated with image data in edge environments to extract human-centered features and characterize them as context. We integrated locally executable image recognition technology based on multiple pre-trained models for human-centered context recognition from live images. Following this approach, an in-home monitoring system can be developed as a standard using a computer and USB camera.