Study algorithmic and machine-learning solutions suitable for detecting the position and orientation of 3D face scans, including methods for automated detection of anthropometric descriptors like landmarks or Frankfurt horizontal. The expected output is a point in space and a vector determining the head pose. The estimated head pose can then be used to accelerate the registration of faces or optimize Poisson disk sub-sampling, for instance.

The text part of the thesis will describe possible methods with a primary focus on approaches selected for implementation. The evaluation chapter will discuss the methodology and results of experiments that aim to evaluate the efficiency and reliability of selected algorithms. The impact of estimated head pose on the efficiency of face registration would provide additional valuable output. The decision to include this evaluation will be based on the complexity of the implementation and evaluation of the estimation algorithms.

Selected pose estimation algorithms will be implemented and integrated into the FIDENTIS Analyst II application. Functionality will be covered by unit tests.

References:

• 3D Face Recognition under Pose Varying Environments, 2004
• Using Facial Symmetry to Handle Pose Variations in Real-World 3D Face Recognition, 2011
• Zhou, S., Xiao, S. 3D face recognition: a survey. 2018: Velmi hezký přehledový text zahrnující ML metody pro "head pose" detection, feature points detection, registration, atd. Obsahuje spoustu dalších užitečných odkazů.