摘要
本文针对在视频追踪过程中出现的目标遮挡问题,提出了一种基于稀疏表达的混合模型的粒子滤波跟踪算法.这种混合模型采用了基于全局模板和基于局部的描述方式,在全局模板的描述方式中,将目标模板由目标候选表示出来,线性表示的系数满足稀疏性约束条件,其系数作为目标候选的权重.同时在局部描述模型中,构造SIFT特征的完备字典,将局部模型稀疏表示成直方图形式,然后对遮挡部分进行处理,设置目标被遮挡部分的直方图权重,得到最终的局部模型直方图表示.最后本文将两种模型合理的融合到一块,得到一种联合的新的模型应用于目标跟踪,实验证明该方法有效的完成了视频中的目标跟踪.
The object visual tracking algorithm based on a mixed model with sparse representation of particle filter is proposed to solve the problem of occlusion.We propose a mixed model that exploits both holistic templates and local representations.In the holistic templates,the target template is represented by a set of all target candidates.The representation coefficients are sparse,and the coefficient is used as weight of the target candidate.Meanwhile,in the local description model,constructing the object dictionary with the image SIFT-features and proposing a novel histogram-based that takes the spatial information of each patch,and dealing with the occlusion handing scheme of the histogram with the weight to get the final local histogram model.At last,we fuse the two modules reasonably get a new model for object tracking.The experiments we conducted prove that this method complete the target tracking effectively.
出处
《天津理工大学学报》
2014年第5期31-35,共5页
Journal of Tianjin University of Technology
基金
国家自然科学基金(61001174)
关键词
混合模型
稀疏表达
遮挡
粒子滤波
mixed model
sparse representation
occlusion
particle filter