采用稀疏和平滑双约束的增量正交映射非负矩阵分解目标跟踪

发布时间：2018-03-17 23:36

本文选题：非负矩阵分解　切入点：稀疏约束　出处：《计算机辅助设计与图形学学报》2017年09期 　论文类型：期刊论文

【摘要】：针对目标跟踪在遮挡和尺度变化等复杂背景下跟踪性能下降问题,联合稀疏约束、时间平滑约束以及增量投影非负矩阵分解,提出一种在线目标跟踪算法.首先利用非负矩阵分解学习一个基于部分表示的子空间,在此基础上添加稀疏约束提高处理遮挡能力,添加时间平滑约束提高算法的稳定性;然后用增量方式完成子空间的在线更新,减少算法计算量、提高外观模型更新效率;最后在粒子滤波框架下,以重构误差为基础改进了观测似然函数,将具有最大后验概率的候选目标作为目标在当前帧的图像区域.实验结果表明,在各种含有遮挡和尺度变化的视频中,该算法可以更稳定地跟踪目标.
[Abstract]:Aiming at the degradation of tracking performance in complex background such as occlusion and scale variation, the tracking performance of target tracking is reduced by combining sparse constraint, time smoothing constraint and incremental projection nonnegative matrix decomposition. An online target tracking algorithm is proposed. Firstly, a subspace based on partial representation is studied by using non-negative matrix decomposition. On this basis, sparse constraints are added to improve the ability of processing occlusion, and time smoothing constraints are added to improve the stability of the algorithm. Then the on-line updating of subspace is accomplished by incremental method, which reduces the computational cost and improves the efficiency of appearance model updating. Finally, based on the reconstruction error, the observational likelihood function is improved under the framework of particle filter. The candidate target with the maximum posterior probability is used as the target in the image region of the current frame. The experimental results show that the algorithm can track the target more stably in all kinds of video with occlusion and scale variation.
【作者单位】：安徽大学计算智能与信号处理教育部重点实验室;安徽大学媒体计算研究所;
【基金】：国家自然科学基金(61372137,61301295) 安徽省自然科学基金(1308085QF100,1408085MF113) 安徽大学信息保障技术协同创新中心开放课题(ADXXBZ201411);安徽大学博士科研启动基金
【分类号】：TP391.41

【相似文献】