摘要
ViBe是一种像素级的背景建模、前景分割算法,其效果优于目前人们所熟知的几种算法,但其还是有检测目标不够完整、因背景与目标颜色相近或光照剧烈变化而容易形成鬼影等不足。针对这两个问题,提出一种结合GrabCut与五帧差分法的ViBe目标检测算法。首先通过ViBe与五帧差分法提取的前景进行“与”运算,然后对所得结果进行形态学处理,减少所得目标分割图像的噪声,并且在非理想状态下,对被分割为多个区域的有效区域进行合并,最后利用GrabCut算法对每个有效区域进行分割。实验结果表明,该算法可有效提取出完整的前景图像,并且减少了背景与目标颜色相似或光照变化对ViBe算法的影响。
ViBe is a pixel level background modeling and foreground segmentation algorithm,and its effect is better than several com⁃mon algorithms.However,it still has some shortcomings,such as incomplete detection target,ghost shadow formed due to the color similarity between the background and the target or the sharp change of illumination.Aming at these problems,we propose a ViBe tar⁃get detection algorithm based on GrabCut and five frame difference.First of all,the foreground extracted by ViBe and five frame differ⁃ence is used for“and”operation,and the results are processed by morphology to reduce the noise of the human shape segmentation im⁃age.In the non ideal state,the effective regions which are divided into multiple regions are merged.Finally,each effective region is segmented by GrabCut algorithm.Experimental results show that the proposed algorithm can effectively extract the whole foreground im⁃age,and reduce the influence of background and target color similarity or illumination change on ViBe algorithm.
作者
熊玖朋
李旭健
潘纪成
XIONG Jiu-peng;LI Xu-jian;PAN Ji-cheng(College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266590,China)
出处
《软件导刊》
2021年第4期106-110,共5页
Software Guide
基金
国家重点研发计划项目(2017YFC0804406)
山东省重点研发计划项目(2016ZDJS02A05)。
