Abstract: This paper proposes a face locating method based on active infrared illumination and Adaboost, which greatly improves performance in robustness and speed. In this system, an active infrared illumination structure is used to get eye candidates, and then face candidates are extracted, followed by a cascaded real Adaboost algorithm to detect all faces. Experimental result shows that it takes no more than 6 ms to locate all faces in one frame for a video stream with a resolution of 640×480.

Key words: autostereoscopic, face detection, bright eye effect, Haar feature

摘要： 针对高分辨率下人脸定位速度慢和鲁棒性差的问题，提出一种鲁棒的基于红外照明的Adaboost人脸定位方法。该方法使用近红外结构照明来得到眼睛的候选点，依据眼在脸部的相对固定位置得到人脸的候选窗口，使用级联的Real Adaboost算法定位出所有的人脸。实验结果显示，这种方法能有效提高定位速度和准确度，对一个640×480分辨率的连续视频，系统每帧人脸定位时间小于6 ms。

关键词: 自由立体显示, 人脸检测, 红眼效应, Haar特征

CLC Number: 

• TP18