摘要
针对视觉背景提取模型在运动目标检测中消除鬼影时间长,采用固定阈值分割前景和背景像素对场景适应性不强,容易出现目标空洞等问题,提出一种改进的视觉背景提取算法.通过三帧差分技术去除运动目标区域,获取真实背景像素初始化背景模型,以消除鬼影.根据背景模型像素的方差及邻域像素的方差,自适应获取每个像素点的分割阈值.实验结果表明,该方法能够去除鬼影,更好地保持目标完整性,提升对场景的适应性.
When visual background extractor（ VIBE） is used to detect moving object the ghost will be kept for a long time and it is not adaptable enough for scene because of the use of fixed threshold which may cause holes in the object,so an improved VIBE model is proposed. Firstly,the moving object region can be detected by the three-frame difference. Then the real background pixels can be obtained to initialize the background model. In this way,the ghost can be removed. In order to improve the adaptation to scene,the threshold of a pixel is determined by the variances of its neighborhood pixels and background model. Experiment results showthat the ghost can be removed and a more integrated object can be obtained and the adaptation to scene is promoted.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第11期2559-2562,共4页
Journal of Chinese Computer Systems
