利用CSK算法从图像碎 片中提取运动目标的一个最小二乘分类, 引入多通道颜色特征标定运动目标, 通过当前图片碎片中的核函数周期性假设循环结构, 一定程度拟补CSK算法使用目标灰度特征描述能力的不足。采用PCA降低特征维度并去除特征冗余信息, 提高分类器参数更新速度, 解决了CSK算法分类器参数更新线性化、无法适应目标发生较大变化时的运动目标跟踪问题。利用benchmark测试平台的算法集与测试数据集进行了实验, 目标颜色核相关跟踪算法(TCKCT) 的实验结果表明, 对光照变化、背景杂乱、目标形变、目标运动速度较快、目标运动幅度较大的情况下, 算法都有较好的跟踪效果。无人机跟踪遥控小车的物理实验结果, 进一步验证了TCKCT算法特性, 良好的实时性能够满足无人机目标跟踪要求, 具有良好的实际应用前景。

The CSK algorithm was used to extract a least square classification of moving objects from image fragments in this paper, and the multichannel color features was introduced to calibrate the moving objects. Through the cyclic hypothesis of periodicity of the kernel function in the current image fragments, the CSK algorithm was applied to compensate the lack of target gray-level features describing capacity with CSK algorithm in some extent. The PCA was used to reduce the feature dimension, remove feature redundant information, improve the updating speed of classifier parameters. The problem of moving target tracking could be solved when CSK algorithm classifier parameters were updated linearly and could not adapt to large changes of target. Experiments were performed on the algorithm dataset of the benchmark test platform and the dataset of test data. The experimental results of target color kernel tracking algorithm (TCKCT) show that the algorithm has a better tracking effect in the case of the illumination changing, the background clutter, the target deformation existing, the target moving velocity is faster and the target motion amplitude is larger. The experimental results of UAV tracking remote control car further verify the characteristics of TCKCT algorithm and good real-time performance can meet the target tracking requirements of UAV. It has a good practical application prospect.