基于连续型Adaboost算法和Cascade结构的红外人脸检测

Infrared face detection based on real Adaboost algorithm and Cascade structure

自由立体显示技术中,人脸位置的探测与跟踪是关键之一。由于光照变化等因素的影响,对多人的脸部位置的探测很难达到快速、准确的目的。提出一种基于连续型Adaboost算法和Cascade结构的新方法。该方法采用红外主动照明模式,通过隔离可见光照,基本消除了光照变化对人脸检测造成的影响。新检测算法中Adaboost检测速度很快,Cascade结构可以检测那些难以识别的人脸,大大地提高了人脸检测的速度和鲁棒性。对视频流图像进行的检测实验中,没有出现人脸"漏检",极少出现非人脸的"误检"。检测速度在Windows XP,Penti-um IV,图片分辨率为640×480的条件下,可达25 f/s,完全达到了实时性的要求。另外,实验证明该方法对于人脸表情变化和人脸小角度倾斜也具有鲁棒性。

Location and tracking the human faces is one of the critical technologies in free stereoscopic display system. But because of illumination variation and some other reasons, it is difficult to detect human faces accurately and fast. In this paper, an infrared face detection based on real Adaboost algorithm and Cascade structure is implemented. With active infrared illumination and separating of visible light, the problem caused by variation of illumination is almost solved. Meanwhile,the combination of real Adaboost and Cascade structure pays more attention to the human faces which is more difficult to identify, making the detection more robust and quicker a lot. In the detection of video sequence, all human faces can be detected, and misdetection rarely appears. The average processing time on a windows XP, PIV 2.4 GHz PC is about 40 ms for a 640 × 480-pixel image. So the improved detection is real-time. In addition, experiment proves that the improved detection is robust when there is variation of facial expression or a little degree leaning of human face.