User-Independent Face Landmark Detection and Tracking for Spatial AR Interaction

We present novel face landmark detection and tracking methods which are independent of user facial differences in a scenario of Spatial Augmented Reality (SAR) interaction. The proposed methods do not require a preliminary general face model to detect or track landmarks. Our contributions include: (i) fast face landmark detection, which is achieved based on our modified Latent Regression Forest (LRF) and (ii) model-independent facial landmark tracking by revising outliers based on a direction and displacement of neighboring landmarks. We also discuss (iii) feature enhancements based on RGB and depth images for supporting several interaction scenarios in SAR environments. We anticipate that the proposed methods promise several interesting scenarios, even under severe head orientation in SAR interaction without wearing any wearable devices.