一种基于块区域统计矩的实时运动检测方法

采用变异系数（Coefficient of Variation, CV）对图像序列帧进行块区域统计矩描述，采用共线性检验（Colinearity Test）的方法，给出了变异系数的最大似然估计（Maximum Likelihood Estimation, MLE）,构建了基于块区域变异系数最大似然比的三帧差分运动检测模型，最后给出了实验结果。实验表明，该模型能够较完整地检测到运动物体的轮廓边缘，且对于光照渐变、阴影及背景物体小幅度扰动等都具有较强的鲁棒性。

A Block-matching Motion Detection Model Based on Statistical Moment

In this paper, coefficient of variation (CV) is used to describe the variations between corresponding blocks of successive frames. Illumination-invariance property of CV is then proved. employing colinearity test algorithm, the MLE（Maximum Likelihood Estimation） of CV is given on frames with additive noise. Based on the maximum likelihood ratio of CV of successive blocks, a novel three-frame differencing model, namely, the MLR model, is then built. Experimental results on real-world video sequences show the effectiveness of the proposed approach, compared to some other traditional motion detection algorithms. This model can usually detect almost the whole silhouette of a moving object, and is robust to the changes of background such as gradual illumination changes, shadow and vacillation etc.