基于自适应阈值Kirsch-LBP纹理特征的均值漂移目标跟踪算法

Mean Shift Object Tracking Algorithm of Adaptive Threshold Kirsch-LBP Texture Features

针对以颜色特征建立概率模型的Mean Shift目标跟踪算法在光线变化时存在的缺陷,提出了一种融合改进型纹理特征与HSV颜色特征的Mean Shift目标跟踪算法。首先,设计一种具有抗光性能的自适应阈值Kirsch-LBP纹理特征算子,该算子利用Kirsch算子的8个方向模板所求的差值,并采用LBP模板均值作为自适应阈值,再按照旋转不变LBP原理提取局部纹理特征;其次,利用不同特征相似性系数间的关系作为加权准则来构建新的权重;最后,将其嵌入到Mean Shift算法中以实现目标跟踪。对比实验结果表明,本算法在光线变化场景中也具有良好的目标跟踪特性,广泛适用于光照变化和姿态变化等复杂场景下的目标跟踪领域。

For improving the performance of the mean shift tracking algorithm based on single color feature, which is adopted to establish the target model when the light changes, a new mean shift object tracking algorithm combining a Kirsch-LBP texture feature with HSV color feature was presented. Firstly, the paper presented a novel adaptive thres- hold Kirsch-LBP feature description operator with light resistance, which uses eight direction difference of the Kirsch operator, uses the LBP template average as an adaptive threshold,and then according to the rotation invariant principle of LBP extracts local texture feature. Secondly, it used the relationship between similarity coefficient of different features as the weighted criteria to construct the new weight. Finally, it was embedded into the mean shift algorithm to realize target tracking. Experimental results show that the algorithm can improve the accuracy of target tracking effectively in the scene of light changing,and improves the performance of traditional mean shift object tracking algorithm.