Abstract: A novel algorithm of face fast detection and real-time tracking that accurately detecting and tracking faces in the video sequences is proposed in this paper.The algorithm contains two parts that are face detection of initial state or object losing state and object tracking of continuous state.Firstly,the central place between two eyes is predicated,and the region around the candidate position is matched with face template,then,the accurate position of the face is detected rapidly.Subsequently,the face template is updated by the detected face region,and the following video images are tracked in a small field based on detected parameters with rotation angle,position and zoom scale.The experimental results on many video under varying conditions show that the face tracking algorithm in the video sequences is fast,accurate and robust.

Key words: face detection, object tracking, template matching, face verification

摘要： 提出一种新的人脸快速检测与实时跟踪算法,能够对视频序列中的人脸进行快速、准确地检测和跟踪。算法分为开始状态、目标丢失状态的人脸检测和连续状态的目标跟踪。首先预测人脸两眼之间的中心位置,得到人脸的预测位置并对预测位置处的图像进行模板匹配,快速检测出人脸准确位置。然后利用检测出的人脸修正人脸模板,并在检测出的位置、旋转度、缩放比例等条件下,对后面序列图像进行小位置、小角度的快速跟踪。实验采用了多种环境下的大量视频,结果显示该算法能够快速跟踪视频序列中的人脸并具有很高的准确性、鲁棒性。

关键词: 人脸检测, 目标跟踪, 模板匹配, 人脸确认