摘要
背景差分法是一种重要的运动目标分割方法,但是其不仅对背景质量的要求较高,且易将运动阴影误检测为前景目标。针对上述问题,提出了一种用于智能交通系统的新的运动目标分割方法。该方法首先在RGB空间对像素灰度归类法进行了改进,并用其提取了背景图像,同时结合选择更新和背景调整来实时更新背景;然后对背景差分图像的RGB灰度之和,通过设定阈值来提取运动区域;最后对提取的目标阴影混合区在HSV空间,分别进行自上向下、自左向右及其反方向的色调、亮度及边界交叉点判别,以实现阴影检测和消除。实验结果表明,该新方法能获得高质量的重构背景,并能消除阴影(尤其是暗色目标的阴影),因此可提高目标的分割质量。
The background difference is important for segmenting mobile objects. But this method highly depends on background quality and easily regards moving shadows as objects. To cope with these problems, a novel segmentation method is proposed for intelligent transportation system. Firstly, the background image is extracted in the RGB space by improving pixel grayscale classification,and is updated real time with selective update and background adjustment. Then, the motion regions are detected by summarizing and thresholding the RGB values in the difference image. Finally, the hue, value and border intersection, which are judged from top to bottom, from left to the right and inverse directions in detected regions, are utilized to detect and eliminate shadows in the HSV space. The experimental results show that this new method can effectively reconstruct background, eliminate shadow （especially dark object shadow） and improve segmentation quality.
出处
《中国图象图形学报》
CSCD
北大核心
2009年第10期2023-2028,共6页
Journal of Image and Graphics
基金
国家自然科学基金项目(60574006
60804017)
江苏省"六大人才高峰"基金项目(SJL-2002-05)
关键词
灰度归类
背景重构
目标分割
边界交叉点
阴影消除
pixel grayscale classification, background reconstruction, object segmentation, border intersection, shadow elimination
