一种对光照变化鲁棒的均值漂移跟踪方法

A Robust Mean Shift Tracking Method Under Varying Illumination

颜色作为一个有效的视觉特征,被广泛的用于基于表面模型的跟踪中。但在跟踪过程中,由于光照、视角及摄像机参数等的变化,往往会造成目标颜色的改变,使得跟踪不稳定。该文提出了一种新的基于局部背景动态修正模糊颜色直方图的均值漂移跟踪方法,在颜色-空间域运用核密度估计建立目标的模糊颜色直方图模型,利用目标的局部背景动态修正目标模型,克服了基于传统颜色直方图建立目标模型时对于光照变化较为敏感的缺点。实验验证了该文算法可以平滑相似性表面,减小局部极值点对跟踪的影响,在光照剧烈变化的情况下能够实时鲁棒地跟踪目标。

Color can provide an efficient visual cue for tracking based on appearance models. However, the apparent color of an object depends upon the illumination conditions, the viewing geometry and the camera parameters, all of which can vary during tracking and therefore make the tracking based on apparent color models unreliable or even failed. In this paper a mean shift tracking algorithm is proposed based on dynamic corrected fuzzy color histogram, which employs local background information around the target to correct the apparent models and overcomes the sensitive of conventional color histogram to illumination change and noise. The algorithm is tested on several image sequences and the results show that it can smooth the similarity surface and achieve robust and reliable frame-rate tracking under varying illumination conditions.