采用稀疏特征选择的红外运动目标跟踪方法

Infrared moving target tracking based on sparse feature selection

复杂场景下的红外运动目标对比度低且缺乏细节信息,难以实现稳定持续跟踪。分析了典型红外运动目标的特性,提出一种稀疏编码与特征选择的改进跟踪算法。采用Logistic回归模型,通过对正负样本的监督学习,计算得到最佳权重特征矢量,并将原始特征模板和粒子采样对象均向该特征矢量投影,削弱了背景成分对运动目标跟踪的影响并降低了运算量。在模板更新策略上采用了每帧更新的方法以适应运动目标的机动性。文中给出的方法与其他两种经典方法的实验比较,证明了本方法对运动目标跟踪的有效性。

The infrared moving target in complex background has the characteristics of low contrast and few details,and it is difficult to realize a stable and continuous tracking. After analyzing the characteristics of infrared moving target,an improved tracking algorithm based on sparse encoding and feature selection is proposed. Using Logistic regression model and the supervised learning of the positive and negative samples,the optimal weight vector was calculated. Then the original feature templates and particle samples were projected to this vector by using a diagonal matrix,which can reduce the effect of cluttered background on moving target tracking and reduce the calculated amount. The update of each frame is used to adapt the moving target maneuverability in template updating strategy. Experimental results show that this algorithm is effective for infrared moving target tracking compared with IVT algorithm and L1 algorithm.