中文摘要：

针对目前目标跟踪的大部分判别算法注重跟踪效率而没有考虑尺度变化这一问题,提出了一个简单而鲁棒的基于颜色统计特征的判别跟踪方法。这种新的颜色统计特征具有一定的光照不变性,同时保持较强的判别能力。建立了跟踪过程的仿射运动模型,利用优化参数来解决尺寸及角度变化等问题。此外,为了进一步提高跟踪速度,采用低维的颜色统计特征描述目标外观,利用颜色统计特征训练贝叶斯分类器,将置信值最大的样本作为跟踪结果,并在线更新分类器。与现有跟踪器的大量综合性的对比实验表明,该判别跟踪方法在不同挑战因素下均有明显优势。

英文摘要：

In view of the problem that most discriminative algorithms for target tracking focus on the tracking speed while neglect the scale variation,a simple,robust discriminative algorithm based on color statistical characteristics was presented. The new color statistical characteristics not only possess the certain illumination invariance,but also maintain the higher discriminative power. An affine kinematics model for tracking was established to keep optimizing the parameters during tracking to solve the scale variation and the view angle change. To improve the tracking speed,the low-dimensional color statistical characteristics were used to describe the target appearance,and the color statistical characteristics were used to train naive Bayes classifiers and update classifiers online. The sample with the maximum confidence was regarded as the tracking result. Numerous comprehensive experiments were conducted for evaluation of the proposed algorithm and other algorithms,and the remarkable effectiveness under different challenge factors of the proposed tracking algorithm was showed.