基于阴影流和3D MAP-MRF的运动阴影消除

Moving Shadow Elimination Based on Shadow Flow and 3D MAP-MRF

阴影消除是运动检测中的一个重要问题。本文提出一种用阴影流和三维马尔可夫随机场后验概率最大化方法运动阴影消除算法。首先对每个像素建立混合高斯模型,通过阴影弱分类器,将可疑的阴影像素分离出来送到阴影流模型中。在线学习候选阴影像素,得到置信度高的阴影流模型。然后用混合高斯模型,阴影流和当前图像一起构建一个三维的马尔可夫随机场模型,将运动目标检测转化为标号组后验概率最大化/能量函数最小化。最后,构建一个与三维MRF对应的三维网络流图,通过动态图切算法,求出图的最小切,即求得MRF标号组的最大后验概率,从而给每个像素分配"前景"或"非前景"标号,达到消除阴影分割运动物体的目的。实验表明本方法在实际视频中取得了较好的效果。

Elimination of shadow is an important issue in moving object detection.In this paper,we present a novel approach of moving shadow elimination based on Shadow Flow and maximum a posteriori probability of 3D Markov Random Field（3D MAP-MRF）. Firstly,Gaussian Mixture Model（GMM）is built as background model of per pixel.By comparing current pixel and GMM,we classify candidate shadow pixel through a shadow weak classifier and send it to Shadow Flow Model.Through on-line learning the candidate shadow which comes from weak classifier,our method get high confidence shadow model.Then,3D MRF is constructed of GMM, Shadow Flow and current images.MAP-MRF/min energy is deviated from moving object detection.Finally 3D graph is constructed according 3D MRF.A dynamic graph cuts algorithm is used to find min-cut/max-flow,which is equal to a maximum posteriori probability of label.Each pixel is assigned by＂foreground＂and＂non-foreground＂label,and moving object detection with shadow elimination is completed.Experiments show that our approach achieves excellent performance.