一种多模型协同的目标提取方法  

An Object Extraction Method of Multi-model Cooperation

作　　者：冀续烨 陈明[1] 冯国富[1] 赵海乐[1] 

机构地区：[1]上海海洋大学信息学院,上海201306

出　　处：《计算机工程》2015年第5期254-258,263,共6页Computer Engineering

基　　金：国家"863"计划基金资助项目(2012AA101905)

摘　　要：固定摄像机目标提取多以高斯混合模型为背景模型,在检测运动缓慢、间歇停滞的目标时会出现前景目标空洞的问题。为此,提出一种能够适应目标间歇停滞的多模型协同目标提取方法。采用高斯混合模型进行背景学习,通过光线检测模型和场景状态检测模型协同控制背景适时更新,利用阴影检测模型剔除阴影。实验结果表明,与Kaew Tra Kul Pong P方法相比,该方法能较完整地提取到目标轮廓,且单帧处理时间较少。Gaussian Mixture Model（ GMM） is adopted to solve foreground detection problems. However,GMM can not detect objects in which do not move in the scene. This paper proposes the multi-model cooperative method to detect foreground objects in complex scene. Under the assumption that the camera is fixed,it first uses the adaptive GMM to build a background which is updated by the light detection model and the scene detection model. A shadow detection model is also used in this paper at last. It mades a comparison with two algorithms. Experimental results show that this method can completely extract the object contour,and single frame processing time is less.

关 键 词：目标提取 高斯混合模型 光线检测模型 场景状态检测模型 阴影检测模型 背景更新 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]