探讨了自动人眼状态分类问题,提出一种基于全局扫描并验证策略的分类框架.该方法采用一种级联结构(Cascade)来组织分类器,采用Adaboost算法学习分类器.实验表明,该方法无论在鲁棒性、正确率和速度方面都达到了很好的性能,具有非常明显的实际应用价值.
In this paper, we discuss the problem of automatic human eye state classification. A new classification framework based on a global scan and verification strategy is proposed. This method employs a cascade structure to organize a series of eye state classifiers trained by Adaboost. Experiments over a large dataset show that the proposed system reaches a good performance in robustness, correctness and speed, and it is of significant value in real applications.