提出了一种基于局部特征索引结构的目标跟踪方法, 将BoVW(视觉词袋)引入到跟踪方法中。构建了一种不依赖于具体特征类型的目标跟踪方法, 较好地解决了实际应用特征描述子在进行相似度度量时的计算和度量问题。再通过对误匹配特征的剔除和未匹配特征的关联预测,提高跟踪的准确性与鲁棒性, 最后对目标区域前景特征进行分离并对目标区域进行最优更新,得到更为精确的跟踪结果。

A target tracking method is proposed based on local feature index structure,which introduces Bag of Visual Word(BoVW) into the tracking method.A target tracking method that does not rely on a specific feature type is constructed,which solves the calculation and measurement problems when using the feature descriptor to perform similarity measurement in actual applications.Then,the accuracy and robustness of tracking are improved by eliminating the mismatched features and the correlation prediction of unmatched features.Finally, the foreground features of the target area are separated and the target area is optimally updated to obtain more accurate tracking results.