融合纹理信息和预测机制的人脸跟踪算法

摘要

针对人脸遇到部分遮挡或类肤色背景时会出现跟踪失败的问题,提出一种连续自适应均值漂移(continuously adaptive mean shift,CamShift)融合边缘特征和卡尔曼滤波预测机制的人脸跟踪算法.主要从两方面进行考虑:利用可变局部边缘模式(varied local edge pattern,VLEP)算子获取目标区域内的人脸边缘特征直方图,结合颜色直方图建立人脸区域特征模型;使用卡尔曼滤波对人脸的空间位置进行预测,获得下一帧图像人脸目标的初始搜索位置.实验结果表明,该算法克服了人脸遇到类肤色背景或部分遮挡时跟踪失败的问题,比传统的CamShift算法具有更鲁棒的跟踪结果.%Aiming at the problem of unsuccessful tracking when the face is partially occluded or in the background similar to skin color,a face tracking algorithm was proposed fusing CamShift,edge texture feature and Kalman filter.Two aspects were considered.One was that face edge texture feature histogram was obtained using varied local edge pattern (VLEP) descriptor,and color histogram was combined to get feature model of face region.The other was that the space position of face was predicted using Kalman filter,and initial searching position of face was obtained.Experimental results show that the improved algorithm overcomes the mentioned problems,and gets more robust tracking results,compared with the traditional CamShift algorithm.