基于模板的非参量统计的视频图像中人的检测跟踪方法

Human Detection in Video Based on Template Non-parametric Statistical Framework

为了克服低分辨率条件下复杂环境和人本身变化对视频图像检测的影响,将人体分为多个部分,通过高斯函数,为每个部分建立颜色线性模型.通过多个线性模型的组合,使得特征更加具有鲁棒性.在Meanshift方法基础上,利用空间位置关系建立模板,得到基于模板的非参量统计方法.该方法要优于单一的Meanshift方法.

In order to detect human in images under lower resolution, complex environment and transformation of human self were overcome. The body of human was divided to several parts which are represented as linear model of color for overcoming these problems. The feature of body could be robust through the composition of these linear models. Based on Mean shift,a template using relative position was combined with Mean shift to get a detecting method which is better than single Mean shift.