一种用于运动目标检测的多模态非参数背景模型

被引:10
作者
毛燕芬
施鹏飞
机构
[1] 上海交通大学图像处理与模式识别研究所
[2] 上海交通大学图像处理与模式识别研究所 上海
[3] 上海
关键词
运动目标检测; 非参数背景模型; 核密度估计; 多样性采样;
D O I
10.16183/j.cnki.jsjtu.2005.s1.031
中图分类号
TP274.4 [];
学科分类号
0804 ; 080401 ; 080402 ; 081002 ; 0835 ;
摘要
提出了一种基于多样性采样原理的高斯核密度估计模型用于多模态背景描述.从包含运动物体的训练序列中,提取具有较高频度以及最大多样性的样本集用于背景建模.并根据新样本及邻域点在总样本集中取值的相关频度计算权值,避免了采用全部训练点产生的信息冗余和重复计算等缺点,使背景核估计的计算简单有效.对复杂场景下车辆监控系统进行实验,结果表明,该算法在提取运动物体中是有效的.
引用
收藏
页码:134 / 137
页数:4
相关论文
共 7 条
[1]  
A probabilistic background model for tracking. Rittscher J,Karo J,Joga S,et al. The 6th European Conference on Computer Vision[C] . 2000
[2]  
Wallflower : Principles and practice of background maintenance. Toyama K,Krumm J,Brumitt,B,et al. International Conference on Computer Vision . 1999
[3]  
Learning patterns of activity using real-time tracking. Stauffer C,Grimson W E L. IEEE Transactions on Pattern Analysis and Machine Intelligence . 2000
[4]  
Non-parametric model for background subtraction. Elgammal A,Harwood D,Davis L. The 6th European Conference on Computer Vision . 2000
[5]  
Detection moving objects, ghosts and shadows in video streams. Cucchiare R,Grana C,Piccardi M,et al. IEEE Transactions on Pattern Analysis and Machine Intelligence . 2003
[6]  
W4: real-time surveillance of people and their activities. Haritaoglu I,Harwood D,Davis L S. IEEE Transactions on Pattern Analysis and Machine Intelligence . 2000
[7]  
Tracking groups of people. McKenna S J,Jabri S,Duric Z,et al. Computer Vision and Image Understanding . 2000