融合局部三值数量和色度信息的均值漂移跟踪

Mean Shift Tracking Method Combing Local Ternary Number with Hue Information

该文提出了局部三值数量(Local Ternary Number,LTN)这一新的局部显著性纹理算子,并将其与色度信息相结合得到一种新的目标跟踪方法。该方法充分利用目标像素与其八邻域像素灰度值的大小关系,将局部显著性算子(Local Similarity Number,LSN)加以拓展,设计了局部三值数量这一新的局部显著性纹理算子,该算子能区分目标像素在同一局部显著度下的不同纹理结构;LTN掩膜提取边缘、线和角点上关键像素以提高纹理特征的区分能力,同时能够较完整地保留目标信息;在此基础上,将掩膜内目标像素的LTN特征与色度信息融合生成一种新的目标模型,并嵌入到均值漂移(Mean Shift,MS)框架完成目标的跟踪。实验结果表明,该文提出的目标跟踪方法在场景中存在相似颜色和光照变化干扰的情况下,仍能持续准确地实现目标的定位,提高了传统均值漂移跟踪算法的性能。

A new local saliency texture descriptor named Local Ternary Number （LTN） is defined in this paper, and a new object tracking method combining LTN with hue information is proposed. Derived from Local Similarity Number （LSN）, the new local saliency texture descriptor makes full use of pixel intensity relationship in a 8-connected neighbourhood. Compared with LSN, the LTN could distinguish different texture structures with the same local saliency degree. LTN mask not only extracts the key object pixels on edges, lines and corners to enhance the distinguish ability of texture feature, but also preserves the unity of the object. Then a new Mean Shift （MS） tracking method is built by utilizing the new object model, which integrating LTN with hue information from the object pixels within this mask. A large number of experiments are conducted and the experiment results demonstrate that the proposed method improve the performance of the traditional MS tracker, and the proposed method tracks object continuously and accurately even in the scene with similar color and illumination changing.