Eecient Color Segmentation under Varying Illumination Conditions

Video-based human body tracking using color segmentation often causes problems due to movements of the object or changes in the global illumination parameters. In this paper the innuence of varying lighting conditions and large object movements in space is measured and analysed. For color segmentation an eecient technique for histogram-classiication is used. Based on the measurement results we propose a simple method to combine the histograms of several measurements which improves the segmentation. The method leads to an automatic calibration procedure which allows stable segmentation under varying lighting conditions. Finally, an application of this method to a model-based person tracking system is described.