Mr. Martin Böckeler finished his Bachelor thesis “Landmark detection in 3D facial data” in Feb. 2013. Detection of facial landmarks like eyes, nose, mouth etc., is an essential step in automatic face recognition. Based on detected landmarks, faces can be aligned to a canonical pose and scale and can then be compared with other references in the database. Landmark detection with 3D facial data is robust to pose and illumination variations in comparison with 2D data. However, 3D detection normally costs more computational time.
In this thesis, a fast and accurate 3D landmark detection algorithm was developed. In the algorithm, gradient images were derived from 3D raw data. The trained classifiers for different landmarks are applied, which exploit Haar like features and consist of several cascaded weak classifiers. Anthropometric constrain was taken into account to limit search space. Even the challenging landmarks such as pupils can be reliably detected. The experimental results shows very promising performance.