Evaluation of head pose estimation methods for a non-cooperative biometric system

摘要

Automatic head pose estimation is playing more and more important role in current computer vision applications. Precise localization of face landmark points together with subsequent analysis allow to determine person's gaze direction or its facial expression. Such solutions are widely used in driver assistance systems, intelligent environments like smart rooms, human-computer interfaces or identification systems. Surprisingly, in the literature, the tasks of determining face yaw or pitch angles and localizing landmark points are approached as separate problems. In this work, authors combine these two issues and analyze them together. Initially, a deep investigation of other researchers work is presented. On this basis, authors examine selected methods in the context of their application in non-cooperative recognition system. Experiments are conducted on a new, dedicated for this purpose multi-pose face dataset. Obtained results yield mean error of face yaw angle calculation of 3°.