Abstract: An adaptive features selecting algorithm using HybridBoost is proposed.Feature ranking classifiers are constructed by utilizing background and object information，and update during the tracking time.Kalman filter is used to predict the motion area roughly，then the targets are tracked by utilizing ranking classifiers and mean-shift algorithm.The experiment result shows that the proposed tracking algorithm can select features adaptively for tracking according to different targets and background information.Even if there exists covering，clutter，appearance and illumination changed in the scene，targets still can track in real time effectively.

Key words: adaptive features selecting, HybridBoost, feature ranking classifier, tracking

摘要： 提出利用混合Boosting算法根据目标信息和背景信息选择特征，建立特征排序分类器，并在跟踪的过程中不断自适应更新。采用卡尔曼滤波对目标区域进行粗预测，然后利用排序分类器结合mean-shift算法完成目标的精确跟踪。实验结果表明，该算法可以根据不同的目标和背景信息，自适应地进行特征选择，对于场景中存在光照、干扰、遮挡等情况，依然可以对目标进行实时有效的跟踪。

关键词: 自适应特征选择, 混合Boost, 特征排序分类器, 跟踪