基于Online AdaBoost的偏振图像目标跟踪

Polarization image object tracking based on Online AdaBoost

根据偏振图像的特点,文章提出一种基于在线AdaBoost的目标跟踪方法。该方法以最小二乘回归作为弱分类器,以强度、偏振度和边缘方向特征组成的向量为其输入;通过AdaBoost算法将多个弱分类器集成为强分类器,并在跟踪过程中利用AdaBoost算法对强分类器进行在线更新,以适应目标与背景的变化;利用强分类器生成当前置信图,在置信图上利用粒子滤波估计目标的状态。实验结果表明,该方法能够在复杂背景下稳定地跟踪目标。

A polarization image object tracking method is proposed based on Online Adaboost. The least square regression is used as the weak classifier, with feature vectors of intensity, polarization degree and margin direction chosen for the input. The ensemble of weak classifiers is integrated into a strong classifier by AdaBoost algorithm, and this strong classifier is updated online by AdaBoost algorithm during tracking in order to adapt to the change of the object or the background. Current confidence map is given by using the strong classifier, and based on this map the new state of the object is estimated by using particle filter. Experimental results show that the proposed method is robust and efficient in polarization image object tracking on complex background.