摘要
与仿射变换相比,投影变换能够更精确地描述目标的成像过程。提出一种基于投影群(SL(3)群)和协方差流形双重建模的视觉目标跟踪算法。算法充分考虑目标动态几何形变和表观的更新,将投影群与协方差矩阵黎曼流形相结合建立了双重的粒子滤波器。一重滤波器用于在投影群上对目标的几何变换参数进行动态更新,另一重滤波器用于在协方差矩阵黎曼流形上在线更新目标的观测模型,2个滤波器交替执行以跟踪动态变化的目标。实验结果表明,所提出的算法优于现有基于仿射变换模型的目标跟踪算法,而且对于目标经历剧烈光照变化或遮挡等条件下,仍具有准确、稳定的跟踪效果。
Compared with affine transformation, projection transformation can describe the object imaging process more accurately. The paper proposes a novel visual object tracking algorithm with dual modeling based on projection group （SL（3）group） and covariance manifold. The algorithm fully considers the dynamic geometric deformation and appearance update of the target, and combines projection group and covariance matrix Riemannian manifold to build two particle filters. One filter is used for dynamic updating geometric transformation parameters of the target on the projection group, the other one is used for online updating the target observation model on covariance matrix Rieman- nian manifold; and the two particle filters are alternately employed to track the dynamic changing target. Extensive experiment results prove that the proposed algorithm is superior to existing object tracking algorithm based on affine transformation model, and can realize stable and accurate tracking not only when the target undergoes significant geo- metric deformation, but also when the target undergoes large illumination change or occlusion.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2014年第2期374-379,共6页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61273078)
辽宁省教育厅科学技术研究项目(L2011212)
