| 利用混合高斯和拓扑结构的人体“鬼影”抑制算法 |
| |
| 引用本文： | 许国梁,周航,袁良友.利用混合高斯和拓扑结构的人体“鬼影”抑制算法[J].智能系统学报,2021,16(2):294-302. |
| |
| 作者姓名： | 许国梁 周航 袁良友 |
| |
| 作者单位： | 北京交通大学 电子信息工程学院，北京 100044 |
| |
| 摘 要： | 若在建模时存在目标，部分目标像素会进入背景模型，会在检测时产生“鬼影”。为了有效抑制“鬼影”，提出一种利用混合高斯和拓扑结构(Gaussian mixture model and topological structure，GMMT)的人体“鬼影”抑制算法。算法分为两个阶段，背景建模阶段采用双通道建模，通道一利用混合高斯模型进行预检测，接着利用拓扑结构将分散的人体目标连接获得完整的目标并取其外接矩形，然后将矩形外的像素加入背景模型，经过多帧的建模得到空背景；通道二使用多帧平均法计算背景模型。通过设置建模帧数的阈值T选择建模方式，若建模帧数小于T则使用通道一建模，否则使用双通道联合建模。目标检测阶段利用改进的背景差分法实现人体分割并进一步消除 “鬼影”。经过测试，GMMT在建模阶段存在目标的情况下可有效地抑制 “鬼影”。
|
| 关 键 词： | 人体检测 背景建模 “鬼影” 混合高斯模型 网状拓扑 均值漂移 背景差分法 像素邻域 |
Human “ghost” suppression algorithm using Gaussian mixture model and topology |
| |
| XU Guoliang,ZHOU Hang,YUAN Liangyou.Human “ghost” suppression algorithm using Gaussian mixture model and topology[J].CAAL Transactions on Intelligent Systems,2021,16(2):294-302. |
| |
| Authors: | XU Guoliang ZHOU Hang YUAN Liangyou |
| |
| Affiliation: | School of Electronic and Information engineering, Beijing Jiaotong University, Beijing 100044, China |
| |
| Abstract: | When modeling, if a target is present, some of its pixels will appear in the background model, which produces a “ghost” during detection. To effectively suppress this “ghost,” we propose a human “ghost” suppression algorithm that uses a Gaussian mixture model and a topological structure (GMMT). The proposed algorithm contains two main stages: a background modeling stage and a target detection stage. In the background modeling stage, the GMMT algorithm adopts double-channel modeling. A Gaussian mixture model is used in channel 1 for pre-detection. Then, scattered human objects are connected by a topological structure to obtain the complete target and its bounding box. Pixels outside the bounding box are added to the background model, and the background is obtained by multi-frame modeling. The multi-frame averaging method is used in channel 2 to calculate the background model. The modeling method is selected by setting the threshold T of the modeling frames. Channel 1 modeling is used when the modeling frame number is less than T, otherwise double-channel joint modeling is used. In the target detection stage, the improved background difference method is used to realize segmentation of the human body and eliminate the “ghost” during modeling. Test results prove that the GMMT algorithm can effectively suppress a “ghost” if a target is present when modeling. |
| |
| Keywords: | human body detection background modeling “ghost” Gaussian mixture model mesh topology Meanshift background difference method pixel neighborhood |
|
| 点击此处可从《智能系统学报》浏览原始摘要信息 |
|
点击此处可从《智能系统学报》下载全文 |
