针对前视红外序列图像中目标与背景对比度低、灰度级动态范围小以及目标边缘模糊的特点,提出了一种基于均值漂移和特征匹配的红外目标鲁棒分层跟踪算法。首先利用均值漂移算法来快速搜索局部最优候选目标。然后,通过特征匹配对跟踪误差进行有效修正。采用Harris算子对模板目标与候选目标进行特征点提取,利用改进的Hausdorff测度对特征点进行有效的匹配度量,最终实现红外目标的精确定位。实验结果表明该算法简单有效,能准确跟踪前视红外目标。

A novel layered object tracking algorithm in FLIR imagery was proposed based on mean shift algorithm and feature matching. First, infrared object was modeled by kernel histogram. Bhattacharyya coefficient was used to measure the similarity between object model and candidate model. The object was then localized by mean shift algorithm rapidly and efficiently. Because of the low contrast between infrared object and background, low dynamic range of gray level, however, the mean shift tracking results may bring some errors. So, feature matching was employed to eliminate the tracking errors. Feature points were extracted in template object and candidate area by Harris detector. Finally, the accurate localization of infrared object was realized by matching the feature points with the measurement of improved Hausdorff distance. Experiment results verify the effectiveness and robustness of this extraction algorithm which can improve the tracking performance efficiently.