Abstract: Considering the features of passenger flow and application requirements of rail transportation, this paper focuses on the global features extraction. It explores the upper features extracting and the Tzone features extracting based on detailed description of image preprocessing technology, facial space establishment and eigenface recognition method. The recognition method based on global features is the identification by extracting the overall facial shape features which is easily affected by expressions, poses, and shades. The recognition method based on the upper features and the Tzone features is proved to be superior to the recognition method based on global features in terms of solving problems of impressions and shades. They are summed according to their own weight. Therefore, the paper proposes a facial recognition method based on features fusion. Pilot application of the proposed method at some stations on BeijingShanghai high speed line proves that this method effectively improves the facial recognition accuracy.

Key words: rail transportation, facial recognition, features fusion, intelligent video, feature extraction

摘要： 结合轨道交通客流特点和应用需求，在详细描述人像识别中的图像预处理、人像空间的建立及特征脸识别方法的基础上，研究了全局特征提取、Upper特征提取和Tzone特征提取方法.基于全局特征的识别方法通过提取人像的整体形状特征进行识别，易受表情、姿势、遮挡等的影响.基于Upper特征和Tzone特征在处理表情和遮挡等问题时，与基于全局特征的方法相比具有一定的优势，因此，本文对全局特征、Upper特征和Tzone特征进行了加权融合，提出一种基于多特征融合的人像识别方法.经京沪高速铁路部分车站试点验证，该方法能有效提高人像识别的准确率.

关键词: 轨道交通, 人像识别, 多特征融合, 智能视频, 特征提取