摘要
提出一种新的基于图模型纹理分割的视频人体检测方法。首先利用帧间拆分与背景拆分相结合的方式对人体进行粗分割;然后确定目标区域,利用高斯图模型建立纹理模型。通过变量选择和参数估计对纹理特征进行分析,并计算纹理图像的最大后验概率,将纹理后验概率大的归到一类。将基于图模型的纹理分割方法应用到视频检测,实验显示了很好的效果。
This paper proposes a new method to detect body in video based on texture segmentation of the graph model. First- ly, interframe split and background separation are used to roughly segment the human body. Then, the target area is deter- minde and the Gauss figure model is used to establish texture model. Through variable selection and parameter estimation the texture features is analyzed, and the maximum posterior probability of texture image is calculated, which makes it easy to ssify the big posterior probability of the texture to a category. Experiments show that the result is satisfactory.
出处
《河北工业科技》
CAS
2012年第3期163-166,共4页
Hebei Journal of Industrial Science and Technology
基金
华北科技学院校立基金资助项目
"中央高校基本科研业务费"资助项目(2011B031)
关键词
高斯图模型
纹理分割
参数估计
视频检测
Gauss model
texture segmentation
parameter estimation
video detection
