Facial Landmark Detection

The computer-vision step that locates anatomical reference points on the face — eye corners, nose, jaw, lip outline — used as the input to Action-Unit detection.

Modern facial analytics typically begins by detecting between 68 and 478 landmarks per face per frame using a deep neural network. Landmarks track the geometry of the face across head poses and partial occlusions.

Action Unit detectors operate on the landmark stream plus the underlying image patches, producing per-frame AU intensity scores that feed downstream emotion classifiers, attention models, and engagement analytics.

Related terms: action unit, emotion ai, affective computing

Back to Facial Analytics Glossary