Abstract: With a brief description of target tracker based on Spatial-Temporal Context（STC） and Circulant Structure with Kernels（CSK）, it mainly introduces the tracker based on Color Name（CN）, and in the light of tracking drift resulting from the occlusion of targets, scale variation or illumination changes, presents an improved method based on adaptive learning rate and adaptive Gauss-kernel scale factor, aiming at improving the training model updating and marking to reduce the cumulative error of the target model and increase the accuracy of the tracking process. In the experiment, it chooses ten types of videos and uses three evaluation parameters to compare the tracking performances between initial and improved algorithms. The results show that the improved algorithm has better robustness for occlusion, illumination changes and scale variation, which is applied to the pedestrian tracking of Unmanned Aerial Vehicle（UAV）, achieving good performance.

Key words: adaptive, learning-rate, Gauss-kernel scale factor, Color Name（CN）-tracker

摘要： 在简要说明基于空时上下文（STC）和基于核函数循环结构（CSK）目标跟踪器的基础上，重点介绍基于颜色特征（CN）的跟踪器，并针对其在目标被遮挡、尺度变化和光照发生变化时易发生跟踪漂移的问题，提出自适应学习速率和自适应高斯核尺度因子两种方法，分别对训练模型的更新和标记进行改进，减少目标模型累积错误，提高跟踪过程准确性。实验中，选取10个视频集，采用3类评价参数对比算法改进前后跟踪效果。实验证明，改进后的算法对遮挡、光照变化和尺度变化等具有较好的鲁棒性，同时将该算法应用在无人机（UAV）视频行人跟踪上，取得了良好效果。

关键词: 自适应, 学习速率, 高斯核尺度因子, 颜色特征（CN）跟踪器