摘要
针对运动目标检测中光照变化、移动阴影问题,提出一种基于多维特征的核密度估计运动目标检测方法。提出一种改进的局部纹理特征二值模式,对噪声和灰度尺度变化具有更好的鲁棒性,在背景建模中将该纹理特征与颜色特征融合进行概率核密度估计,并结合像素的邻域相关性抑制虚假前景以更好地应对多模态背景场景。实验结果表明:在基于纹理特征或核密度估计的同一体系算法中,本文方法对光线渐变以及运动柔性阴影都有较好的鲁棒性,综合性能指标提高了18%;与目前性能优越的算法纵向比较,能在平均检测性能相当的情况下提高50%的处理速度,更好地平衡检测效果与时间性能。
To solve the problems of illumination change and moving shadow in video moving target detection,a new Kernel Density Estimation(KDE)method is proposed based on multi-dimension characteristics.An improved local texture feature binary pattern is put forward which is robust to noise and variant gray scale.In background modeling,the local texture feature and color feature are fused for KDE,and the neighborhood correlation is integrated to suppress false foreground.Experimental results show that the proposed method has good robustness to slow illumination change and soft project shadows and enhances F_measure by 18% comparing to the same system algorithms based on local texture feature or KDE model.Comparing to existing state-of-the-art methods,the proposed method can enhance processing speed by 50% while maintaining the same detection result.The method has comprehensive performance,which balances detection effect and time cost.
作者
金静
党建武
翟凤文
王阳萍
申东
张振海
JIN Jing;DANG Jian-wu;ZHAI Feng-wen;WXNG Yang-ping;SHEN Dong;ZHANG Zhen hai(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;Gansu Provincial Engineering Research Center for Artificial Intelligence and Gruphics & Image Processing,Gansu Computing Center,Lanzhou 730070, China)
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2019年第2期647-655,共9页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(61562057)
甘肃省高等学校科研项目(2017D-08)
甘肃省科技计划项目(17JR5RA097)
兰州市人才创新创业项目(2015-RC-8)
兰州交通大学校青年基金项目(2015003)
关键词
信息处理技术
运动目标检测
核密度估计
局部纹理特征
邻域相关性
information processing technology
moving targets detection
kernel density estimation
local texture feature
neighborhood correlation
