一种改进的基于纹理和颜色的运动阴影检测

精确地消除活动阴影对运动目标的影响是智能视频监控的核心任务之一,针对当前运动阴影检测中采用的纹理信息过于粗糙、阈值选取需要人工干涉等问题,通过对NCC(归一化互相关)纹理算法进行改进,并结合亮度和归一化颜色特性,提出一种自适应的运动阴影检测方法。以混合高斯模型得到的前景像素为基础,通过阴影在亮度和归一化颜色的特性筛选出候选的阴影区域,结合改进的纹理算法进一步缩小阴影区域范围,最后利用空间后处理得到真实阴影。实验结果表明,该算法在有效降低噪声干扰的情况下能够较好区分局部纹理不明显的运动目标和阴影。

An improved moving cast shadow detection based on texture and color

Accurate shadow elimination for video moving objects is an important issue in intelligent video monitoring. As the texture information applied in current moving cast shadow detection is excessively rough and the threshold is not automatical, a novel adaptive elimination algorithm based on (Normalized Cross-correlation),intensity and normalized color is proposed. In this scheme,the foreground pixels are obtained using Gaussian mixture model,the candidate shadow region is selected through intensity and normalized color feature,and the texture featuer difference is adopted to get the more accurate shadow region, finally,the spatial analysis is performed to confirm the true shadows. Experimental results confirm the validity of the algorithm,and indicate that the moving object and the shadow without obvious local texture could be well separated.