关键词: 粒子滤波, 均值漂移, 目标跟踪, 多特征

Abstract:

Used in video sequences, an algorithm for multifeature particle filter (PF) is put forward with resampling. The new CamShift is applied to each particle in order to make all particles reach a stable position nearby. Likelihood function for each particle is computed by color and edge Bhattacharyya coefficients. The method exploits the CamShift characteristic of low computational complexity, which makes PF realize target tracking by less particles than before. Analysis shows that the method is robust to occlusion.

Key words: particle filter, meanShift, target tracking, multiple features