复杂光照环境下视频序列人脸姿态跟踪方法研究  

Face pose tracking in video sequences under complex illumination

作　　者：余小兰 燕敏 YU Xiaolan;YAN Min(Chengdu Jincheng University,Chengdu 610000,China)

机构地区：[1]成都锦城学院,成都610000

出　　处：《激光杂志》2022年第6期201-205,共5页Laser Journal

基　　金：四川省高等教育学会“十二五”高等教育科学研究规划项目(No.SC125YB069)。

摘　　要：复杂光照环境下难以实现人脸姿态的高精准跟踪,为此研究复杂光照环境下视频序列人脸姿态跟踪方法,提升跟踪进度和效率。通过高频加强滤波和直方图均衡技术光照补偿视频序列图像,采用参考颜色表法匹配图像肤色序列特征,将匹配到的图像肤色序列特征作为待分类样本输入Boosting分类器,利用改进Adaboost算法自动挑选特征,检测待跟踪人脸姿态特征,以此为基础,通过创建二维肤色高斯模型,并不断更新肤色模型,克服复杂光照条件对肤色带来的影响,实现复杂光照环境下人脸姿态跟踪。实验结果表明,该方法对于平移、缩放、旋转、光照及遮挡等复杂光照环境下的人脸姿态都能较好跟踪,在摄像机固定状态下,跟踪准确率为98.57%,计算时间为1.5 s,在摄像机运动状态下,跟踪准确率为93.33%,计算时间为1.86 s,抗噪声干扰性能较为优越。It is difficult to achieve high-precision face pose tracking in complex lighting environment. Therefore, the face pose tracking method of video sequence in complex lighting environment is studied to improve the tracking progress and efficiency. High frequency enhanced filtering and histogram equalization technology are used to compensate the illumination of video sequence images. The reference color table method is used to match the skin color sequence features. The matched skin color sequence features are input into boosting classifier as the samples to be classified. The improved AdaBoost algorithm is used to automatically select features and detect the face pose features to be tracked. Based on this, a two-dimensional skin color image is created Gaussian model, and constantly update the skin color model to overcome the influence of complex lighting conditions on skin color, and realize face pose tracking in complex lighting environment. The experimental results show that this method can track the face pose well in complex illumination environments such as translation, scaling, rotation, illumination and occlusion. The tracking accuracy is 98.57% and the calculation time is 1.5 s when the camera is fixed, and the tracking accuracy is 93.33% and the calculation time is 1.86 s when the camera is moving, so it has superior anti-noise interference performance.

关 键 词：复杂光照 视频序列 人脸 姿态跟踪 分类器 肤色特征 

分 类 号：TN91[电子电信—通信与信息系统]