基于时域相关性的背景建模算法

BACKGROUND MODELING ALGORITHM BASED ON TEMPORAL CORRELATION

针对监控视频前景存在较强的时域相关性降低背景建模性能的问题,提出一种基于时域相关性的背景建模算法。该算法有以下两个创新点:提出LDBCBR(Long Distance Block Composed Background Reference)算法,使用背景搜索间隔IBBS来削弱前景的时域相关性,生成更纯净的背景模型;提出二次建模算法,将LDBCBR和BCBR结合使用,当它们在同一位置都搜索到临时背景块时,将搜索到的临时背景块进行二次建模得到最终背景图像。实验结果表明,算法能够建模生成更纯净的背景图像,与BCBR算法和基准档次对比获得的BD-Rate增益分别为3.12%和25.70%,对监控视频编码有很大提升。

Aiming at the problem that the foreground of surveillance video has strong temporal correlation,which reduces the performance of background modeling,this paper proposed a background modeling algorithm based on temporal correlation. The algorithm has the following two innovations. Firstly,we proposed a long distance block composed background reference(LDBCBR) algorithm,which used background search interval IBBS to weaken the temporal correlation of foreground and generated a cleaner background model. Secondly,we proposed a secondary modeling algorithm,which combined LDBCBR and BCBR. When they searched for temporary background blocks in the same location,the searched temporary background blocks were re-modeled to get the final background image. The experimental results show that the algorithm can model and generate more pure background images. Compared with BCBR algorithm and baseline profile,the BD-Rate gains are 3.12% and 25.70% respectively,which greatly improves the video coding for surveillance.