摘要： 传统智能监控系统在阳光下进行测试时，阴影对检测结果影响较大。为此，提出一种用于视频的阴影去除算法，该算法用阴影的光亮、RGB和色差等物理特性得到初步去除阴影的目标结果，运用形态学方法扩充和完善目标，在获得目标的真实边缘和膨胀方向基础上，利用灰度、纹理和梯度的相似性对目标进行扩充。实验结果表明，该算法的平均正确率从68.47%提高到89.17%。

关键词: 阴影检测, 阴影去除, 目标跟踪, 智能监控系统, 高斯混合模型, 边缘检测, 图像亮度, 图像梯度

Abstract: In intelligent surveillance system, the result is seriously influenced by shadow on sunny days. This paper proposes an algorithm to remove the shadow from images. It detects shadow regions using physical features of shadow for an initial result with less shadow. Three properties of shadow, luminance, RGB and chrominance are taken into account, to get the primary result. And uses morphological methods to improve result. Using the similarity of gray density, texture value and the gradient value, a good result can be obtained. Experimental result shows that the general accuracy of this algorithm is increased from 68.47% to 89.17%.

Key words: shadow detection, shadow removal, object tracking, intelligent surveillance system, Gaussian Mixture Model(GMM), contour detection, image luminance, image gradient

中图分类号: 

• TP18