Abstract

Cloud-based elderly monitoring systems create surveillance risks with limited data protection enforcement. This study validates technical feasibility and cost-effectiveness of edge-based pose estimation for elderly safety monitoring in resource-constrained contexts, demonstrating how privacy principles can drive architectural decisions from inception. The proposed architecture comprises four 850nm NIR cameras and an NVIDIA Jetson Orin Nano edge platform. Privacy protection emerges from system constraints: immediate conversion of video frames to skeletal coordinates, pose-only storage, and permanent video deletion. The integrated YOLOv8n and MediaPipe pipeline achieves 100% on-device processing. Validation on 20 commercial CCTV videos demonstrates 91.3% keypoint detection at 20.53 FPS, confirming 24/7 monitoring capability without facial recognition technology. Edge architecture reduces 3-year total cost by 61% compared to cloud alternatives, expanding market accessibility to 168,000–252,000 elderly individuals in middle-income Cambodian urban households. This work demonstrates that governance principles can drive technical architecture in healthcare AI for Cambodia.