基于非参数核密度模型的交通图像目标提取

Target extraction in city traffic image based on nonparametric kernel density model

针对现有目标提取和去噪方法不能很好地满足城市交通图像车辆目标提取的要求,提出基于概率比较结合形态学闭操作的目标提取去噪方法。通过非参核密度估计算法建立背景模型,获得每个像素点上各灰度值的出现概率,提取出前景目标;分别计算前景目标是属于车辆移动还是树叶抖动的概率,通过概率比较去除噪声,用形态学闭操作进一步去噪。实验结果表明,提出的算法较好地实现了树叶噪声与车辆目标的分离,能有效去除树叶抖动噪声,正确提取车辆目标,具有良好的抗噪性。

Based on the probabilities analysis and mathematical morphology operation,a new traffic target extraction and denoising algorithm is proposed.Nonparametric kernel density estimation is employed to build the background model and extract the foreground object by getting the probabilities of gray level on each pixel.The probabilities of foreground object are calculated to distinguish whether it is caused by the motion of vehicles or the fluttering of the leaves and the noise is removed by the comparison of the probabilities.The treated image is denoised further by the applying of mathematical morphology.The experiment results show that the algorithm can effectively separate vehicle target from noises,remove the noises caused by the fluttering of leaves,and extract the target correctly with good noise proof feature.