一种基于肤色与LBP特征融合的人脸跟踪算法

Face Tracking Algorithm Based on Color and LBP Feature Fusion

针对目前人脸跟踪方法易受光照变化和背景相近色的干扰,跟踪效果有时不佳或失效的问题,提出引入LBP(Local binary pattern)局部纹理特征,采用LBP直方图和颜色直方图相融合作为人脸特征描述的粒子滤波人脸跟踪方法。该方法在全局颜色和局部LBP纹理两个层次和特征线索上对人脸进行描述。实验结果表明,该方法较单一特征跟踪方法更具鲁棒性。此外,由于人脸目标的运动通常为非匀速运动,为了提高粒子传播的有效性和指导性,本文对人脸跟踪状态方程进行了改进。实验证明,改进后的人脸跟踪算法在各种复杂背景、旋转遮挡和人脸目标非匀速运动的情况下均能取得较好的跟踪效果。

Due to the poor tracking performance resulted from the single color feature, a new face tracking algorithm is proposed. The algorithm introduces the local binary pattern （LBP） texture feature, and fuses the LBP histogram and the color histogram as face tracking features, then the face is modeled at two different levels and clues by joints of color and local LBP textures. Experimental results show that the algorithm is more robust than that based on the single feature. In addition, according to the non-uniform motion of the face, the system equation is improved to enhance the effectiveness and the guidance of the particle propagation. The algorithm can achieve a better tracking performance under the conditions of the complex background, the rotation, the occlusion, and the non-uniform motion of the face.