颜色属性跟踪算法使用的核脊回归分类器对于背景信息的利用过少, 导致算法在目标快速运动、局部遮挡和背景相似物干扰等情形下容易发生漂移。针对此问题, 首先, 通过计算响应矩阵峰值旁瓣比确定干扰峰所在位置; 然后, 在核脊回归分类器上引入对应的背景约束项来加强分类器对背景信息的利用; 最后, 使用构建尺度空间滤波器的方法实现对多尺度变化的支持。在Visual Tracker Benchmark数据集上的对比实验结果表明, 该方法在不过多损失算法速度的前提下, 能够有较高的跟踪精度。

The kernel ridge regression classifier used in adaptive color attribute tracker makes little use of information about the context, which may easily result in drifting in cases of fast target motion, partial occlusion or background clutter. To solve the problem, the interference peak position is determined at first by calculating the peak side lobe ratio of the response matrix. Then, the corresponding context constraints are introduced on the kernel ridge regression classifier to enhance the background utilization. Finally, a scale space filter is established for supporting multi-scale variation. The experimental results based on the Visual Tracker Benchmark dataset show that this method can achieve high tracking accuracy with a mild loss of frame rates.