Useful urgency of dynamic faces of face to face high method of sorting mode of tracking face-speft co-ax / watanabe

https://www.youtube.com/watch?v=wovjyhbbnt8

Dynamic facial projection projection mapping (DFPM) overlays images made by the computer’s faces to generate harmful experiences used in the makeup and entertainment industry. In this study, we recommend two concepts to reduce artificial artifacts between expected images and basic challenges for DFPM.

Our first concept is a method of tracking high speed tracking tracking factors temporal. We first introduce a crupped-area of ​​limited dismissal of disabling face inter / extrapolation face, allowing the resemblance to kill landmark landmark. We immediately suggest a novel hybrid landmark landmarking technique united in running regression trees (ERT) -MASED detection. ERT-based detales rapidly produce results in 0.107 MS using temporal information with auxiliary arrival support to regain detection errors. To train the facial landmarking method, we recommend a new method for simulating high-frame-rated advertisements available public data available.

Our second concept is a Setup of Co-Axial Projegor-Camera that continues to a long optical alignment of 1.274-pixel error between 1 m and 2 m depth. This setup can reduce abuse by applying the same optical designs to the projector and camera without reason to misorignment as usual ways.

Based on these concepts, we develop a novel high-speed DFPM system achieving almost perfect person’s view.

reference

• Hao-Lun Peng, Kengo Sato, Soran Nakagawa, Yoshihiro Watanabe: Perceptually-aligned Dynamic Facial Projection by High-Speed ​​Face-Tracking Method and Lens-Shift Co-Axial Setup, IEEE TRANSACTIONS ON Visualization and Computer Graphics, Doi: 10.1109 / TVCG.2025.3527203 (2025)