改进型LTP纹理颜色直方图用于非刚体跟踪

Modified LTP Texture Color Histogram for Non-rigid Tracking

由于传统Mean-Shift算法中单一颜色特征描述目标不完整,而现有算法中多特征融合维度大,不能较好地解决非刚体目标跟踪过程中目标形状快速变化的问题。因此,提出了改进型局部三值模式(Local Ternary Patterns,LTP)纹理颜色联合直方图的跟踪算法。该算法首先提取原始LTP纹理特征的关键纹理模式,再采用或运算的方法进行二次合并以降低特征融合的维度,在合并纹理模式形成的掩摸下,将纹理与颜色特征联合并有效地引入到Mean-Shift算法框架,实现目标的准确跟踪。实验结果表明,所提算法能够实现背景干扰、快速运动等复杂场景下的非刚体运动跟踪。与其他算法相比,跟踪误差更小,鲁棒性更强。

For the incomplete target description by the single-color features in traditional Mean-Shift algorithm and the large dimension of multi-feature fusion in the existing algorithms,the rapid change of target shape in non-rigid target tracking process could be fairly solved.Therefore,the tracking algorithm of modified LTP（Local Ternary Pattern） texture and color histogram is proposed.The key texture pattern of the original LTP texture features is firstly extracted in the algorithm,and then the secondary merging method is used to reduce the multi-feature fusion dimensions by using the OR operation.The texture and color feature are combined under the masking of combined texture pattern and effectively introduced into the Mean-Shift algorithm framework,thus to realize the accurate tracking of the target.The experimental results indicate that the proposed algorithm can realize the non-rigid motion tracking in complex scenes such as background interference and fast motion.Compared with other algorithms,this algorithm is smaller in tracking error and stronger in robustness.