本文提出了一种复杂条件下基于子空间梯度方向直方图跟踪的方法,通过大量样本的离线训练构建目标的投影子空间,并用梯度方向直方图在子空间的投影作为新的目标描述特征. 为了满足实时性的要求,采用积分直方图方法

提高粒子特征的计算速度 然后结合粒子滤波方法在子空间中计算粒子与训练样本集之间的相似度,进而估计目标的运动参数.-A subspace t racking method is proposed to t rack target s under complex environment s. First ,

target is represented by subspace histogram of oriented gradient descr iptor , which is comp uted by

projecting histograms of oriented gradient descr iptor to a subspace , t hat is built f rom a large set of

t raining samples before t racking. Then an integral histogram met hod is incorporated to reduce the

comp utational complexity. By t hese , t he t racking problem is formulated as a state inference problem

under t he particle filter f ramework of target motion parameter estimation.