Evaluation Protocols and Validation for Cameras in Indoor Healthcare Monitoring

This study addresses the lack of standardized validation for camera-based monitoring in healthcare. The authors propose two technical validation protocols to evaluate five cameras (four RGB-D and one RGB) in indoor settings. The first protocol assesses metrological performance, including depth accuracy, thermal stability, and field of view. The second protocol evaluates application-level performance by measuring the accuracy of 3D human pose estimation using state-of-the-art algorithms (RTMO and YOLO26) against a gold-standard motion capture system. The experiments systematically varied lighting, camera height, viewing angle, and occlusion levels to simulate real-world deployment challenges.

• Metrological Variability: Depth measurement bias at a 5-meter distance varied drastically between devices, ranging from 50 mm to over 1400 mm.
• Pose Estimation Performance: All tested cameras achieved comparable 2D pose estimation accuracy (mAP between 78% and 90%). However, 3D reconstruction accuracy was highly dependent on the camera's depth-sensing quality, with significant differences in Mean Per Joint Position Error (MPJPE).
• Environmental Impact: While camera mounting height had minimal impact on performance within the tested range, lighting conditions and occlusions had camera- and estimator-dependent effects on 3D reconstruction quality.

For researchers and clinicians, this paper provides a much-needed evidence-based framework for selecting hardware. It highlights that choosing a camera based solely on 2D pose estimation performance is insufficient for applications requiring 3D spatial accuracy, such as gait analysis or fall detection. The study emphasizes that technical validation must account for both sensor-level metrology and application-level algorithmic performance to ensure reproducibility in clinical or home-based monitoring.