一种视频中基于检测的快速人脸跟踪方法

针对背景复杂多变的视频人脸实时跟踪问题,从检测率、漏检率与错检率3个指标出发,通过改进样本选取的方式对训练样本的比例进行优化,得到一种快速人脸检测方法.在对卡尔曼的初始状态进行选取后,运用具有通过调整参数就能对被跟踪的人脸区域进行放大或者缩小功能的卡尔曼滤波方法来选取候选人脸区域,从而减少搜索时间,提高脸部区域检测速度...

A face tracking method based on detection in video

It has proved that the efficiency of training sample can be optimized by improving the selecting sample way, base on three indexes as detection rate, false-negative rate and false alarm rate in this paper. The method of face quick detection consists of two parts training and detection. To use the method for face tracking is Kalman filter, with the setting function by adjusting parameter to reduce or enlarge the region that is being tracked in order to minimize the search time and improve the speed of detection, after selection of initial state of Kalman filter. This method can accomplish our goal of tracking function based on detection. Finally, AdaBoost face detection method based on the improved selecting samples and selecting method based on Kalman filter respectively, and combing with each other. The practical result indicates the method can tracking faces in two videos with good effect.