融合多特征基于图割实现视频遮挡区域检测  

Fusing Multi-feature for Video Occlusion Region Detection Based on Graph Cut

作　　者：张世辉[1,2] 何欢[1] 孔令富[1,2] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]河北省计算机虚拟技术与系统集成重点实验室,河北秦皇岛066004

出　　处：《光学学报》2015年第4期234-243,共10页Acta Optica Sinica

基　　金：国家自然科学基金(61379065);河北省自然科学基金(F2014203119)

摘　　要：为了准确检测视频中的遮挡区域,提出一种融合多特征基于图割的视频遮挡区域检测方法。基于光流和亮度信息提出三种新的遮挡相关特征—亮度块匹配特征、最大光流差特征和光流残差特征,并定义了所提特征的计算方法。以像素点为单位将所提特征组成特征向量输入随机森林分类器,获取像素点及邻接像素点对的遮挡相关信息。综合利用所获取的遮挡相关信息,通过构造遮挡检测能量函数将遮挡检测问题转化为优化问题。根据该能量函数构造无向图,并基于图割理论对能量函数进行求解,从而得到最终的遮挡区域检测结果。实验结果表明,同现有表现较好的遮挡检测方法相比,所提方法具有较高的准确性和较好的实时性。To detect the occlusion region in video accurately, an occlusion region detection approach is proposed for video by fusing multi-feature based on graph cut. Three new occlusion related features named brightness patch match, maximal flow difference and flow residual are proposed based on the information of optical flow and brightness, meanwhile their calculation methods are defined. The feature vector of each pixel is composed of the proposed features and is inputted into the random forest classifier to obtain the occlusion related information about pixels and adjacent pixel pairs. An occlusion detection energy function, which transforms the occlusion detection problem as an optimization one, is constructed by synthesizing the above occlusion related information. An undirected graph is constructed according to the energy function, then the energy function is solved by graph cut theory to gain the final occlusion region detection result. The experimental results show that, compared with the existing advanced methods, the proposed approach has higher accuracy and better real-time performance.

关 键 词：机器视觉 视频序列 遮挡区域检测 光流 随机森林 能量函数 图割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]