摘要
针对实时视频中的多姿态人脸检测问题,应用扩展的类Haar特征,训练能有效检测多种姿态和多种旋转角度人脸的分类器;并使用该分类器实现了一个实时视频的多姿态人脸检测系统。该系统分为训练和检测两个子系统,训练系统应用大量包含正反例子的图片进行训练,得到分类器;检测系统首先使用DirectShow从USB摄像头获取图像,然后读入分类器,对图像进行检测并显示。实验结果表明,该系统能够快速准确地在视频中检测出多种姿态的人脸,有较强的实用价值。
In order to fast detect the multi-view face in Real-time video, the classifier which can effectively detect a wide range gesture and rotation face by using the extended Haar-like features, is trained. With using the classifier, a multi-view real-time video face detection system is built. The system includes two sub-systems, training system and detecting system. Training system uses a large number of images containing both positive and negative examples for training the classifier. In detection system, DirectSbow is used to get real- time images from the USB camera, then images are detected by using the classifier. Experiments show that the system can quickly and accurately detected multi-view face in a real-time video, which makes it more practicable.
出处
《计算机工程与设计》
CSCD
北大核心
2010年第18期4065-4067,4096,共4页
Computer Engineering and Design
基金
江苏大学高级专业人才科研启动基金项目(05JDG020)
