电子学报 ›› 2019, Vol. 47 ›› Issue (7): 1401-1407.DOI: 10.3969/j.issn.0372-2112.2019.07.002

• 学术论文 • 上一篇    下一篇

一种抗光照突变的前景提取方法研究

徐贵力1, 谢瑒1, 程月华1, 王正盛2, 张泽宏1, 姜斌1   

  1. 1. 南京航空航天大学自动化学院, 江苏南京 211100;
    2. 南京航空航天大学理学院, 江苏南京 211100
  • 收稿日期:2018-04-04 修回日期:2019-01-14 出版日期:2019-07-25 发布日期:2019-07-25
  • 作者简介:徐贵力 男,1972年4月生于黑龙江佳木斯,教授、博士生导师.现任南京航空航天大学自动化学院测试系主任,研究方向为计算机视觉与智能系统、光电检测与自动化装置等.E-mail:guilixu2002@163.com;谢瑒 男,1994年7月生于重庆梁平.现为南京航空航天大学自动化学院测试系硕士研究生,研究方向为智能交通、视觉SLAM.E-mail:xieyang3552@163.com
  • 基金资助:
    国家自然科学基金(No.61473148)

A Foreground Extraction Algorithm for Sudden Illumination Changes

XU Gui-li1, XIE Yang1, CHENG Yue-hua1, WANG Zheng-sheng2, ZHANG Ze-hong1, JIANG Bin1   

  1. 1. College of Automation Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu 211100, China;
    2. College of Science, Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu 211100, China
  • Received:2018-04-04 Revised:2019-01-14 Online:2019-07-25 Published:2019-07-25

摘要: 针对经典前景提取算法无法在光照突变情况下正确提取前景的问题,根据LBP算子对光照不敏感的特性,提出了一种基于截尾均值的纹理特征提取算法,即通过对噪声的抑制及对平坦区域序列的稳定性处理,解决了原有LBP算子易受噪声干扰,平坦区域序列不稳定及得到的纹理图信息冗余的问题.结合高质量纹理特征,根据纹理特征的光照不变性,设计了一种能有效应对光照突变情况的背景更新模型,实验结果表明,本文提出的融合纹理特征的前景提取模型不仅能够在光照缓慢变化的情况下有效地对运动目标前景进行提取,而且在光照突变情况下仍然能够进行准确提取,前景提取的准确率相比平均背景模型提高61.7%,相比混合高斯模型提高59.3%.

关键词: 计算机视觉, 前景提取, 光照变化, 纹理特征

Abstract: In this paper we address the problem of foreground extraction from images where there is an abrupt change in illumination.This condition is not adequately handled by classical foreground extraction algorithms;thus,we propose a novel algorithm based on the censoring mean that relies on the LBP (Local Binary Pattern) operator's insensitivity to illumination.Our approach first solves issues related to the stability of the region sequence.In turn,this handles the problems of the original LBP operator being susceptible to noise interference,as well as the instability of the flat region sequence.We have implemented a background updating model that is based on texture invariance,and can effectively deal with abrupt changes in illumination.The experimental results show that our proposed method for the extraction of fusion texture features can handle both slow changes in light,as well as changes in the foreground due to moving objects.The accuracy of foreground extraction can still be improved under the condition of light mutation.Our method performs favorably when judged against the average background model,where,the accuracy of foreground extraction is increased by 61.7% and 59.3% compared to the mixed Gaussian model.

Key words: computer vision, foreground extraction, illumination changes, texture feature

中图分类号: