增量学习Rao-Blackwellized粒子滤波的跟踪算法

Tracking algorithm using incremental learning Rao-Blackwellized particle filter

基于特征子空间的目标跟踪方法能适应目标状态的变化,并对光照等外部环境不敏感,但通常假定特征子空间的基向量固定,这样不仅需要离线训练,而且在目标姿态发生较大改变时,跟踪精度会降低。提出一种基于增量学习的Rao-Blackwellized粒子滤波算法,通过在线学习获得特征子空间的基向量,并用解析的方法对目标在子空间的投影参数进行在线更新。实验表明,新算法在目标有较大形变、姿态变化和光照等条件变化时,能保持较高跟踪精度,具有较强的鲁棒性。

The eigen subspace based tracking method is adaptive to the change of object state and is robust to lighting variation.Usually it supposes the eigenbasis vectors are static and trained offline,so tracking precision will degrade under large pose variation.This paper proposes an incremental learning based tracking algorithm using Rao-Blackwellized particle filter.The eigen- basis vectors in subspace are trained online and the object＇s projection parameter in subspace is updated online.Experiments show that the proposed method is more precise and robust under conditions such as large appearance variation,pose variation and lighting variation.