一种提升鬼影抑制性能的改进视觉背景提取算法  

An Improved Visual Background Extractor for Suppression of Ghosts

作　　者：杜瑾 苏雨 张义飞[1,2] 邹坤 DU Jin;SU Yu;ZHANG Yifei;ZOU Kun(China Airborne Missile Academy,Luoyang 471000,China;National Key Laboratory of Air-based Information Perception and Fusion,Luoyang 471000,China;Northwestern Polytechnical University,Xi'an 710072,China;AVIC Jonhon Optronics Technology Co.Ltd,Luoyang 471000,China)

机构地区：[1]中国空空导弹研究院,河南洛阳471000 [2]空基信息感知与融合全国重点实验室,河南洛阳471000 [3]西北工业大学,陕西西安710072 [4]中航光电科技股份有限公司,河南洛阳471000

出　　处：《郑州航空工业管理学院学报》2024年第1期97-105,共9页Journal of Zhengzhou University of Aeronautics

摘　　要：运动目标检测为视频帧生成指示运动目标像素的二值图,因此,正确判别待检测场景中的运动目标像素和背景像素是运动目标检测算法的主要任务。本文针对传统场景提取算法背景模型初始化导致的“鬼影”问题,提出了时域区间参考模块,该模块通过时域区间内像素的统计值生成背景参考值,在像素判别阶段利用候选前景像素与该参考值的差异进一步确定前景像素。实验结果表明,本文提出的方法对传统视觉背景提取算法性能进行了提升,对“鬼影”有较强的抑制作用,提高了检测准确度,具有较好的检测性能。Moving object detection generates a binary map describing moving object pixels for a frame,so it is the main task of the moving object detection algorithm to correctly discriminate foreground moving pixels from background pixels.To address the "ghosting" problem caused by the initialized background model of the classical visual background extraction algorithm,a time-domain interval reference module is introduced.The module generates a background reference value from the statistics of the pixels in the interval,and then uses the difference between the candidate pixel and the reference value to further determine the foreground pixel.The experimental results show that the algorithm in this paper makes up for the shortcomings of the original visual background extractor,has a strong suppression effect on“ghost”,improves the detection accuracy,and has a better performance for detecting moving object.

关 键 词：运动目标检测 背景减除 鬼影 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]