基于GMM的视频图像人数统计算法

A people counting method based on Gauss mixture model

针对摄像机俯视拍摄场景的人数统计问题,提出一种运算效率高、误检率低的人数统计方法。以人头部位为检测对象,采用运动侦测、边缘检测方法获取人头轮廓,在此基础上采用高斯混合模型(Gauss mixed model,GMM)分别对人头轮廓目标点集和椭圆模型进行建模,通过最小化人头轮廓目标点集与椭圆模型的GMM之间的欧氏距离求解椭圆参数,统计满足椭圆形状的轮廓数量,再通过形状滤波得到人数统计结果。人数统计对比实验结果表明,新方法的误检率低,且运算效率高。

People counting method with high operation efficiency and low false detection rate is proposed to count the people in the top views shoot by camera. Human heads are taken as the detection objects and head sketches are obtained by motion detection and edge detection based on the Gauss Mixture Model which is a- dopted to make the modeling of target point set and ellipse model of human head sketch. Elliptic parameter is solved by minimizing the Euclidean distance between the Gauss Mixture Model of target point set and ellipse model of human head sketch. Then, the number of elliptic sketches is counted. Finally, the number of people is obtained by shape filter. The result of contrast experiments shows that the proposed method has low false de- tection rate and high operation efficiency.