二颜色空间中CamShift人脸跟踪  

CamShift algorithm of face tracking in two color spaces

作　　者：王俊南 赖惠成[1] WANG Junnan;LAI Huicheng(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《现代电子技术》2018年第21期37-41,46,共6页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(61561048);新疆维吾尔自治区科学基金资助项目(2015211C257)~~

摘　　要：针对CamShift算法在对人脸跟踪时容易受到肤色背景的影响,且当跟踪环境中背景较为复杂时易失败的问题,提出一种二颜色空间中CamShift人脸跟踪算法。该算法首先在YCgCr与YCgCb空间中进行肤色分割,将每一视频帧中的肤色区域与非肤色区域有效地分隔开来;然后对保留的肤色区域进行形态学操作,并在经过形态学处理所得的肤色区域中引入人脸几何特征,规定只有当肤色区域的长宽比为0.5~1.0或肤色区域面积大于5 000像素的部分方为人脸;最后使用CamShift算法在规定的肤色区域中跟踪人脸。同时为了自动检测目标人脸以初始化搜索窗口,减少算法的处理时间,提高算法的性能,又引入改进的AdaBoost算法检测人脸。实验结果显示,该算法可以获得有效且可靠的跟踪结果。Since the CamShift algorithm used for face tracking is easily affected by skin color background,and is easy to fail while the tracking area has complex background in tracking environment,a CamShift algorithm of face tracking in two color spaces is proposed.The skin color is segmented in YCgCr and YCgCb color spaces to separate the skin region from non-skin region of each video frame effectively.The morphological operation is performed for the preserved skin color area,and then the facial geometrical characteristic is introduced into the skin color area processed by morphology.If the conditions that the length-width ratio of the skin color area is within 0.5~1.0 or the pixels of skin color area are more than 5 000 is satisfied,this area is determined as the face.The CamShift algorithm is used to track the face in the specified skin color area.An improved AdaBoost algorithm is introduced to automatically detect the target face for search window initialization,reduce the algorithm processing time,and improve the algorithm performance.The experimental results show that the algorithm can obtain effective and reliable tracking results.

关 键 词：YCGCR颜色空间 YCgCb颜色空间 肤色分割 ADABOOST算法 CAMSHIFT算法 人脸跟踪 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]