JMIR Publications

ABSTRACT

Background:

Elderly activity recognition is a critical task in long-term care monitoring, yet it remains challenging due to postural deformities and health-related variability. These factors cause different activities to appear visually similar, or the same activity to appear dissimilar, undermining the effectiveness of traditional human activity recognition (HAR) models developed for the general population.

Objective:

This study aims to improve elderly activity recognition by incorporating care assessment information to enable personalized and context-aware monitoring in real-world care environments.

Methods:

We propose a care-assessment-aware spatio-temporal transformer (CSTT) model that integrates body keypoints, heatmaps, and care level data. The model dynamically adjusts its attention mechanism based on care level context to improve recognition accuracy. CSTT was trained and validated on a real-world elderly motion dataset collected from 28 participants during natural, privacy-preserving mealtime sessions without external intervention.

Results:

Despite data imbalance and considerable intra-class variation due to differing care needs, the proposed CSTT model achieved an F1 score and accuracy of 0.96. Analysis revealed that movement patterns differ significantly across care levels and that similar motions occur in distinct activities, highlighting the importance of care-aware modeling.

Conclusions:

Incorporating care level information into activity recognition models significantly enhances performance in elderly care settings. The proposed CSTT framework demonstrates the value of personalized, context-sensitive approaches for accurate and ethical monitoring in long-term care environments.