中文摘要：

针对复杂背景下视觉目标跟踪的鲁棒性和准确性问题，提出了一种最小均方差意义下的置信图自适应融合跟踪方法。将多种视觉特征统一到置信图表示框架下，并根据各个置信图的分类能力，以均方差最小为评价准则，得到了融合的置信图并用于跟踪。方法兼顾了特征的分类能力及多特征之间的互补性，有效实现了多种视觉特征的融合。实验结果表明，相对于现有方法，能够获得更好的鲁棒性和更高的准确度。

英文摘要：

A novel visual tracking approach based on adaptive fusion of confidence map by minimizing mean square error is pro- posed to improve robustness and accuracy of tracking. In this approach, visual features are assimilated to confidence maps and fused for tracking by minimizing mean square error. This fusing approach combines classification performance and diversity of different features, and achieves better robustness and accuracy than current approaches.