中文摘要：

在线目标跟踪是计算机视觉领域的一个具有挑战性的问题．提出了一种基于特征分组的在线目标跟踪算法．首先，利用像素点在多帧的方差对模板库中的目标模板进行特征分组．然后，利用主要特征图像和次要特征图像学习投影矩阵P，对样本进行投影．最后，利用最小误差法得出当前帧的跟踪结果．与其他典型算法相比，该算法对目标的异常变化具有很强的鲁棒性．

英文摘要：

Online object tracking is a challenging issue in computer vision. An online object tracking algorithm based on feature grouping is proposed. Firstly, the object templates in template base are grouped by computing the variance of the pixels in multiple frames. Then, the projection matrix P is learned based on the more discriminative image and the less discriminative image, and the samples are projected. Finally, tracking results of the current frame are performed by minimum error. Compared with other popular methods, the proposed method has strong robustness to abnormal changes.