基于在线分级特征选择机制的目标跟踪算法

为了提高在复杂背景、光照突变等因素干扰下视频序列目标跟踪算法的鲁棒性,提出了一种新的基于在线目标特征选择机制的跟踪算法。利用目标颜色信息和梯度方向直方图构造目标特征空间,用G auss混合模型对目标特征集建模,分级选择机制在目标特征集中依据最大化信噪比准则选择最优的特征子集,生成概率权重图像。利用概率权重图像构造观测似然函数,在粒子滤波的框架下,实现目标的跟踪。实验结果表明:该算法可以有效地克服目标旋转、背景复杂、光照突变等因素的干扰。

Online cascaded adaptive feature selection for object tracking

An online cascaded adaptive feature selection method was developed to track object in video sequences in the presence of cluttered backgrounds and sudden illumination changes.The algorithm fuses color and histogram gradient features into the tracker and models the object feature sets using a Gaussion mixture model.A cascaded selection strategy based on maximizing a signal noise ratio benchmark is used to select the feature subset.Then a merged probability weight image is generated to calculate the likelihoo...