动态场景中基于绝对差值和的前景检测算法

摘要

针对动态场景中前景检测误报率较高的问题,结合视频序列中背景图像存在时域高冗余的特点,提出一种改进的前景检测算法.基于滑动窗口机制组建微视频基元(MVBE),采用时空绝对差值和计算MVBE的冗余性,对冗余性较低的MVBE采用动态纹理(DT)进行建模.利用批量主成分分析方法估计DT参数,并引入观测性测量方法计算MVBE中每个像素点的显著度以实现前景检测.实验结果表明,该算法相比KDE,GMM等检测算法对场景的动态变化具有更好的抑制效果,并且取得了较鲁棒的前景检测结果.%In order to reduce the high false alarm rate on foreground detection in dynamic scenes,this paper proposes an improved foreground detection algorithm,considering the high temporal redundancy of background images in the video sequence.Firstly,the Micro Video Block Element(MVBE) groups are formed by the sliding window technique and the redundancy values of each MVBE are computed by Spatio-Temporal Sum of Absolute Difference(STSAD) method.Secondly,The MVBE groups of low redundancy are modeled by Dynamic Texture(DT),and the DT parameters are estimated by Batch-Principal Component Analysis(Batch-PCA).Thirdly,the saliency value of every pixel in the MVBE is calculated by the observability measure method to realize foreground detection.Experimental results show,compared with the KDE,GMM algorithm,the algorithm has better inhibitory effect in dynamic scenes and a robust foreground detection result is obtained.