基于混合高斯模型的运动目标检测算法

Moving object detection algorithm based on Gaussian mixture model

针对摄像机在静止条件下的自适应运动目标检测,提出一种改进的运动目标检测算法。首先,针对高斯混合背景建模初期背景建模效果不理想的问题,利用统计的方法得到背景模型,根据背景图像建立高斯混合模型;在模型学习方面,为均值与方差设置了不同的学习率。针对传统的LBP算子的缺陷,提出了一种改进的纹理特征算子,将其与HSV颜色空间去阴影的方法相结合,从而实现对阴影的检测与去除,利用随机Hough算子对圆的检测原理,在运动目标检测的基础之上,实现对人头的边缘检测。实验结果表明:该算法可以很好地检测出运动目标,并能够有效去除运动目标包含的阴影区域,从而实现人头区域的检测。

Under the static condition of a camera for adaptive moving target detection,this paper puts forward an improved algorithm for moving object detection. First of all,considering that in the early stage of Gaussian mixture background modeling,the background modeling effect is not ideal,the background model is obtained by statistical method at the beginning of the video sequence,and then Gaussian mixture models are set up for the background image; then,in aspect of the model learning,different rates of learning are set for the mean and variance in order to improve the convergence rate of the background model. In view of the defects of the traditional LBP operator,an improved texture feature operator is proposed. This improved operator is combined with the method of removing shadow area of the HSV color space,thereby to detect and get rid of the shadow,and further to achieve detection of the edge of human head according to the principle of random Hough operator＇s detection of the ring. The experimental results show that the proposed algorithm can well detect moving targets,and can effectively remove the shadow in the moving object and thereby to achieve the detection of head area.