Abstract: To detect objects quickly，a new method is presented to construct a cascade of linear classifiers with L-SVM（Lagrangian Support Vector Machine，L-SVM）.At first，the negative data is divided into several parts accumulatively according to the geometric distribution of the training data.Here，every part of negative data is separable with the positive data；Second，L-SVM is used to obtain the linear classifiers between every part of negative data and positive data；At last，the linear classifiers are combined to construct a cascade detector.The experiments show that this method enjoys good generalization capacity and much fast speed compared with the traditional SVMs.

Key words: cascade, geometry, Lagrangian Support Vector Machine（LSVM）

摘要： 为了实现目标的快速检测，提出了一种新的基于拉格朗日支持向量机（L-SVM）的线性级联式分类器的构造方法。该方法首先根据样本的几何分布，用迭代的方式把负样本分成若干部分与正样本线性可分的样本；然后用L-SVM对这些正负样本进行分类，得到若干个线性分类器；最后，将这些线性分类器顺次组合，构成级联分类器。实验表明，与经典非线性SVM分类器相比，这种分类器在保持SVM较强泛化性能的优点的同时，在检测效率方面更是具有明显的优势。

关键词: 级联, 几何, 拉格朗日支持向量机