基于暗通道去雾的核相关滤波目标跟踪算法

Object tracking algorithm based on kernelized correlation filter with dark channel defogging

雾霾天气进行目标跟踪时,会出现拍摄到的图像对比度和可见度低的情况,已有的跟踪算法会因为图像特征不明显而出现跟踪漂移甚至导致跟踪失败.针对这一问题,提出一种将核相关滤波与暗通道去雾算法相结合的雾天视频跟踪算法,首先根据目标区域暗通道图的平均灰度值判断是否需要去雾,对需要去雾的图像进行暗通道去雾处理;然后通过核相关滤波对目标进行定位和跟踪;最后根据跟踪结果用去雾后的样本更新分类器.实验结果表明,该算法在大雾情况下可以取得很好的跟踪结果.

The image captured in the foggy weather has low contrast and visibility.The existing tracking algorithm may cause tracking drift even failure because the image features are not obvious.To solve this problem,a foggy video tracking algorithm combining kernel correlation filter(KCF)and dark channel defogging algorithm is proposed.Firstly,it is determined according to the dark channel image grayscale value of object area whether the image needs to be defogged.The image with dense fog is defogged by dark channel defogging.In the second stage,the target is positioned and tracked through kernel correlation filter.Finally the classifier is updated according to the tracking result with samples after defog.The experimental results show that the proposed algorithm can achieve good results in the case of heavy fog.