Research On Key Technologies Of Robust Face Tracking System

Fully automatic face tracking has always received much interest in the computer vision community since it could easily find its way into applications such as visual surveillance, net meeting, video retrieval, virtual reality, auto-focus and stabilizing of imaging devices, etc. This thesis focused on providing an automatic human face tracking system for image sequences. The key work of this thesis consists of several separated parts:Firstly, human face might appear in various environments with different illumination conditions. Hence, we proposed a two-stage based illumination normalization method to remove both vague and hard shadows from a single image. This preprocess phase is instrumental in locating faces under poor illumination conditions.In the second place, the author presented a real-time robust face detector based on the AdaBoost and proposed a novel set of features for boosting. The cascaded detector is rapid and has a high accuracy.The main challenge of tracking is to cope with the appearance variability of target faces. We proposed a novel method to handle large changes in appearance based on online real-value boosting, which is utilized to...