基于多特征融合GMM的阴影检测策略研究

Research of shadow detection strategy based on multi-feature fusion GMM

将运动目标从背景中准确分割出的过程中,阴影的检测和消除起着重要作用。传统的目标和阴影检测算法一般都是基于目标颜色等单特征信息,因此在很大程度上受到了场景光照条件变化的影响而导致算法的执行效果降低。提出一种基于颜色信息和纹理信息的多特征融合的混合高斯模型检测算法,可以降低由单特征检测所带来较高的误检率。其中采用了两重阴影判决方法以确定真实阴影,首先通过颜色夹角进行疑似阴影的判决,进而根据前景区域和背景区域的相似度和颜色分量差值再次判决阴影。最后通过实验对阴影检测算法进行比较,表明了本文提出算法能够对阴影进行准确消除。

During the process of accurately segregating an object from background,shadow detection and removal plays an important role.The traditional target and shadow detection algorithms are generally based on single feature information（for example color）,which is greatly affected by the change of scene illumination condition,resulting in the decrease of implementation effect of the algorithm.In this paper,a color information and texture information based multi-feature fusion GMM background modelling is proposed to reduce the false detection rate caused by single feature.A double shadow judgment method is proposed to determine the true shadow.The suspected shadow is firstly determined by the color angle,then the shadow is judged again according to the similarity and the deviation of color components between the shadow region and the background.Different shadow detection algorithms are compared,which shows that the double shadow judgment method works well in accurate shadow removal.