基于邻域相关性的背景重构

A Background Reconstruction Algorithm Based on Neighboring Correlation

背景差法是一种重要的运动检测方法,当场景中背景被长时遮挡,并非总是以最大频率出现时,往往容易将前景错误地认为是背景,从而产生错误的重构,针对该问题本文提出了一种基于邻域相关性的背景重构算法。算法首先对输入数据进行排序;其次利用简单归类算法对排序后的数据进行分类;再次计算灰度类的出现频率,根据灰度类的出现频率得到像素的背景确定标识,并为背景不确定的像素选定候选背景;最后对背景不确定的像素循环执行背景确定程序,即基于像素邻域相关性,选择和邻域相似度最大的候选背景为像素的确定背景。为验证算法的有效性,对算法进行了多种场景的仿真实验,仿真实验表明,即使在背景被长时遮挡的环境中,算法仍能很好构建背景,从而有利于后续的运动目标检测、识别和跟踪。

The background subtraction is an important method for detecting the moving objects, which is widely applied in the video monitor system. When background is occupied by a foreground for a long time, the foreground will be mistakenly regarded as a background. In order to solve the background reconstruction in which background does not always appears with the largest appearance frequency, a new background reconstruction algorithm based on neighboring correlation is proposed. At first, the data is sorted in an ascending or descending order; secondly, the sorted data is classified by the simple method; thirdly, the appearance frequency of classified classes is computed. The definite identity of background is obtained by appearance frequency. The candidate backgrounds are selected for the pixels without definite background; finally, background selection procedure based on neighboring correlation is repeatedly executed to the pixels until the background of all pixels has been obtained. Simulations results show that the algorithm is able to deal with the complex scene in which the background has been covered for a long time. The proposed algorithm is able to reconstruct the background of scene well, and therefore the target could be perfectly extracted and successfully tracked.