Moving Shadow Detection Based on Multi-feature Analysis and Gray-scale Morphological Filtering

摘要

For many applications in video surveillance systems, shadow detection is a very important step for enhancing the accuracy of moving object segmentation. This paper proposes a robust multi-feature approach which makes integrated use of intensity, chromacity and texture for shadow elimination. After motion detection, foreground pixels are evaluated by several weak classifiers in a coarse-to-fine manner. In each level, the outputs of the weak classifiers are accumulated and formed as an evidence map. Then shadows are eliminated using a gray-scale morphological filtering. Results on various real surveillance videos show the effectiveness and robustness of the algorithm.