基于多信息融合自适应粒子滤波的目标跟踪算法

Object tracking based on multi-information and auto-adapted particle filter

粒子滤波是一种可在非线性和非高斯情况下进行状态估计的有效方法,基于单特征粒子滤波跟踪算法在目标跟踪中的应用及存在的不足,针对颜色信息在光照变化和相似背景条件下存在的缺点,为了仍能对目标进行有效地跟踪,加入纹理信息来表示目标,并给出基于目标的颜色特征和纹理特征多信息融合的自适应粒子滤波算法,使用三种量测模型并给出具体的算法和实验结果,实验证明此方法与仅基于颜色粒子滤波跟踪方法相比,在计算量增加不多的情况下大大改善了跟踪的性能和鲁棒性。

Particle filter is an effective way to estimate the state under non-linear and non-Gaussian. In this paper, based on the shortcomings of characteristics of the particle filter with a single cue and the light of information for color change and conditions similar to the background, in order to track target effectively, texture information is also added to show the target, and we give au- to-adapted particle filter algorithm based on the color characteristic and the texture characteristic multi-information fusion. Using three kinds of measure models, we also give specific algorithms and experimental results. Compared with color-based particle filter tracking methods, the method of this paper improves the track performance and robustness greatly.