具有动态级联结构的在线Boosting算法

针对在线Boosting由于提前设定弱分类器个数导致分类器的分类复杂度受到限制的缺陷,提出了一种新的具有动态级联结构的在线Boosting算法.该算法不但具有一般级联结构的特点,而且能根据输入样本分类的难度来实时地调整级联结构的层数,从而使得目标检测器在检测速度和检测精度方面达到很好的平衡.与一般的方法不同,该算法并没有记录一段短暂的历史样本片段来确定弱分类器的阈值,而是把每一个弱分类器的输出值视为一个随机变量,从而进一步估计它的密度函数.然后以迭代的方式估计出整个强分类器的密度函数,进而构建出在线Boosting的动态级联结构.实验结果表明:与原始的在线Boosting算法相比,该算法大大提高了目标检测的速度和精度.

On-line boosting algorithm with dynamic cascade structure

One drawback of the on-line boosting algorithm is that the discriminative complexity of the classifier is limited since the number of its weak classifiers is fixed in advance.This paper proposes a new on-line boosting algorithm with dynamic cascade structure.It automatically adjusts the number of weak classifiers in real time according to the classification complexity of input sample,which makes the balance of the speed and the precision of objection detection well.Different from usual methods,the algorithm...