一种对光照具有鲁棒性的人脸跟踪算法  

A Robust Face Tracking algorithm Under Varying Illumination

作　　者：袁宝峰[1] 吴乐华[1] 曾伟[1] 吴志强[1] 

机构地区：[1]重庆通信学院信号与信息处理实验室,重庆400035

出　　处：《信号处理》2010年第1期132-136,共5页Journal of Signal Processing

摘　　要：在经典Mean Shift跟踪算法框架中,颜色是一种有效的视觉特征。但在实际跟踪过程中,光照、角度、摄像机位置等的变化会极大地削弱颜色特征的有效性,从而造成跟踪的不稳定。本文提出了一种对光照变化具有鲁棒性的MeanShift人脸跟踪算法。首先,对光照对目标特征域的影响进行建模分析,然后根据分析结果将高斯截断核函数引入到目标概率密度函数中以减小光照变化对目标跟踪的影响,最后将新的目标概率密度函数纳入到Mean Shift跟踪算法框架中。实验仿真结果表明,本文算法在光照剧烈变化的情况下,对人脸的跟踪具有很好的鲁棒性。Under the framework of classical Mean Shift tracking algorithm, color is an effective vision characteristic. But in practice ,the validity of color will be seriously weakened in practice because of the variety of the illumination ,the angle and the location of the vidicon, Consequently, the tracking will be instability. In this paper, a robust Mean Shift face tracking algorithm under varying illumination is proposed. First, modeling and analyzing the effect of illumination, then according to the results of the analysis, bring the truncated Gaussian kernel function into the probability density function to reduce the impact of the illumination. At last, bring the new probability density function into the framework of the Mean Shift tracking algorithm. And the result of the emluator indicated that it can achieve a robust face tracking under varying illumination conditions.

关 键 词：Mean SHIFT 高斯截断核函数 局部二值模式(LBP) 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]