一种基于色彩统计的快速前景检测算法

A FAST FOREGROUND DETECTION ALGORITHM BASED ON COLOUR STATISTICS

针对当前流行的码本模型背景建模和混合高斯模型背景建模对背景像素的分布描述不够精确,建模过程较为复杂,以及处理高分辨率的视频不具备实时性这些问题提出了一种快速简洁的背景建模方法。通过对动态的背景像素在RGB空间的三个色彩分量的统计和分析,发现三个分量相互间的差值在一个狭窄的区域内波动,基于这一事实提出基于RGB色彩分量统计的背景建模方法。该方法充分考虑了RGB三个颜色分量的相关性,较之基于混合高斯模型和码本模型的这两种方法,背景建模的过程更加简单,并且前景检测的过程也更加快捷。实验结果表明该方法不仅能够更准确地描述背景像素的RGB色彩分布,具有良好的鲁棒性,并且大大减少了前景检测的计算复杂度,时间消耗和内存消耗。

For solving the problems of the popular background modelling algorithms of codebook model and Gaussian mixture model thatthey cannot accurately describe the distribution of background pixels,their processes of modelling are both rather complex,and they all lackthe real-time property when dealing with high-resolution pictures,we present a fast and concise background modelling algorithm.Through thestatistics and analysis of the dynamic background pixels in three colour components of RGB space,we notice that the difference between anytwo of them three fluctuates in a very small range.So based on the fact,we propose a background modelling method which is based on RGBcolour components statistics.This method takes full account of the correlation between RGB’s three colour components.Compared with thealgorithms based on GMMmodel and coodbook model,its background modelling process is simpler,and the foreground detection process isquicker as well.The results of experiments show that this algorithm can give more accurate description on the distribution of background pixelsin RGB colour space with better robustness,and can greatly reduce the computation complexity,time cost and memory consumption of fore-ground detection as well.