Dynamic Facial Motion Analysis Using High-Precision Geometry From a High-Speed Metric 3-D Camera System

This work proposes a metric 3D-based concept for dynamic facial motion analysis and presents its metrological validation using a high-speed metric 3D camera system with a color camera and 3D unit composed of GOBO (GOes Before Optics) projector at 850 nm and two calibrated near infrared cameras at the same wavelength. The concept aims to shift facial expression analysis from inference-based estimation toward direct high precision geometric measurement by integrating synchronized color and metric 3D data. After calibration and image registration, 478 standardized facial landmarks are extracted via MediaPipe and transformed into metric 3D coordinates. Based on these landmarks, as proof of concept, 17 geometric features corresponding to selected facial action units (AUs) are defined to provide an interpretable, measurement-driven representation of some facial muscle activations. A systematic uncertainty evaluation framework with different-level metrics, inspired by the GUM and ISO 5725 standards, is established to assess the reliability and repeatability of the measured geometric features. Experimental validation on 10 participants demonstrates that the used metric 3D camera system can precisely measure the facial geometric features with average combined uncertainties of 0.78 mm for distance features and 1.55° for angular features. Under dynamic conditions, measured facial motion induced geometric feature variations are much greater than the uncertainty of measurement, exhibiting strong statistical significance. These results confirm the feasibility of the proposed metric 3D-based concept for reliable and interpretable dynamic facial motion analysis, with potential applications in scenarios like micro-expression recognition, lie detection, and healthcare monitoring.