一种新的多特性联合阴影检测方法

Novel Shadow Detection Method by Integrating Multiple Features

针对运动阴影检测时单一阴影特征难以完全将前景和阴影正确分离,提出一种多特征联合运动阴影检测方法。考虑运动阴影的光照、色度、纹理和区域统计特性,提出采用小邻域光照的对数比值不变性来判定阴影,接着联合阴影HSV颜色空间特性和梯度方向小块合并的阴影区域统计特性来实现多特征联合运动阴影检测。为了客观评价方法性能,采用一种改进的量化方法,对不同光照和环境条件下的视频序列进行测试。实验结果表明,该方法效果好,前景检测率和阴影检测率高,可应用于智能视频监控的目标检测。

To improve the segmentation performance, a novel approach for shadow detection integrating multiple features was proposed, which considers information of color, shading, texture, neighborhoods and temporal consistency to detect shadows in a scene. Firstly, illumination logarithm invariability of neighborhood shadow pixel was proposed to detect shadow. Then, integrating the shadow feature of HSV color space and the statistical feature in a region with the combined blocks based on a gradient algorithm, the shadow was detected efficiently and reliably. Finally, an improved quantitative method was introduced to evaluate the algorithm on a bench mark suite of different illumination and environment video sequences. The experimental results show the effective performance of the algorithm. The method can be applied to moving target segmentation in intelligent video surveillance.