摘要
传统的运动目标检测算法主要基于像素值的统计模型,对于光照突变和噪声极为敏感。为此,提出了一种基于时空LBP建模的自适应运动目标检测算法。通过使用结合了时序信息的LBP描述视频图像序列中像素特征,通过经典的高斯混合模型对像素特征进行建模,提取出运动目标。实验结果表明,该算法能够适应光照变化,具有良好的检测性能。
Traditional moving object detecting algorithms are mainly based on statistical model by pixel illumination,which are extremely sensitive to illumination variance and noises.To resolve this problem,a novel adaptive moving object detecting method is proposed by model pixels with its temporal-spatial local binary pattern(LBP).Pixel features in video image sequences are conducted by temporal-spatial LBP,and then moving objects are extracted by traditional Gaussian mixture model.The results of experiments on I2R databases demonstrate that the proposed algorithm has achieved excellent detection accuracy on complex scenarios.
出处
《四川兵工学报》
CAS
2013年第5期130-133,共4页
Journal of Sichuan Ordnance
基金
安徽省自然科学基金项目(11040606M150)
安徽省高校教学研究重点项目(20101689)
安徽省高校自然科学研究重点项目(KJ2011A048)
关键词
目标提取
光照变化
时空LBP
高斯混合模型
背景更新
object extraction
illumination variance
temporal-spatial LBP
Gaussian mixture model
model update