A study for stabilization of human region detection in outdoor scenes

在这项研究中，我们开发了一种基于样本的方法来有效地检测颜色与背景相似的运动对象。样本数据库是由退化和完整的人体轮廓模式的夫妇。在我们的方法中，退化模式和完整的人体轮廓模式之间的关系制定和退化区域的补偿是通过优化在基于块的方式进行。实验结果证明了该方法的有效性。

In this research, we have developed an exemplar-based approach to efficiently detect moving object that are similar in color to the background. Exemplar database is constructed from couples of degraded and complete human silhouette patterns. In our approach, the relationship between the degraded pattern and the complete human silhouette patterns is formulated and compensation of degraded area is performed by optimization in a block-based manner. The experimental results proved the efficiency of the proposed method.