一类基于信息融合的粒子滤波跟踪算法

Visual tracking using particle filters based on data fusion

本文提出了一种基于图像多特征信息融合的粒子滤波跟踪算法。该算法利用颜色柱状图描述运动目标颜色分布信息,帧间差的梯度图像描述目标运动信息,并在柱状图框架下给出了运动目标颜色和运动似然模型,保证了颜色和运动似然模型在尺度上的统一。由于图像多特征提供了运动目标多方面的测量信息,从而提高了算法的可靠性。试验表明该算法在使用相同粒子数目的情况下较采用单一颜色特征的粒子滤波跟踪算法效果好。

An efficient tracking algorithm based on image multiple cues fusion within particle filter framework was proposed. The proposed algorithm utilized gradient image of frame difference to capture the motion information and color histogram to capture the color distribution information of target, and then color and motion likelihood models within histogram-based framework were developed for the analysis of color and motion cues, which ensured a similar order of magnitude for the two visual cues. The application of multi-cue supplied more information than single cue, which ensured tracking reliability. The experiments on real image sequences show that our tracker with multiple cues has more reliable performance than tracker using particle filters only based on color cue.