Tracking Faces in Grayscale Video Sequences with Binary Direction Vectors - Introduction and Evaluation of Binary Direction Vectors as a New Local Structural Feature for Tracking Faces with the OpenCV CAMSHIFT Tracker

This diploma thesis introduces a novel color- independent feature for image analysis. Furthermore it describes an application prototype using this feature for face-tracking and its extensive evaluation. The main achievement of this thesis is the development of an alternative simple and robust basis for face-tracking solutions and other image processing purposes. This novel method allows the encoding of local structural features that are recognizeable in gray-scale images as so-called Binary Direction Vectors (BDVs). This representation of structural information is successfully combined with the existing tracking algorithm OpenCV CAMSHIFT Tracker to demonstrate the simple handling of BDVs. The tracking precision of the CAMSHIFT/BDV combination is increased by modifying the statistical analysis that is used by the tracking algorithm. The supremacy of the modified version of the tracking algorithm over the original version is proved with an extensive empirical evaluation. These evaluation series also demonstrate how tracking systems can be compared in a precise and scientifically founded way.