基于大数据安全技术及深度特征的鲁棒视觉跟踪

Robust visual tracking based on large data security technology and depth characteristics

针对光照变化、目标旋转、背景杂乱等复杂条件下,核相关滤波KCF算法出现目标跟踪漂移或者失败的问题,本文利用卷积神经网络(CNN)对跟踪目标出现光照、旋转、背景杂乱等复杂变化极具鲁棒性的特点,提出了一种基于卷积神经网络的鲁棒视觉跟踪算法CKCF。CKCF算法在考虑大数据安全和隐私保护技术的前提下,利用海量的图片数据集训练VGG模型提取目标深度特征,并融入改进后的KCF跟踪算法中,实验结果表明,与KCF算法相比较,该算法实现了更加鲁棒的跟踪效果,解决了KCF跟踪算法在光照变化、目标旋转、背景杂乱等复杂条件下目标跟踪漂移或者失败的问题。

According to the problem of target tracking drift or failure for nuclear related filter KCF algorithm under the complicated conditions such as the illumination changes,background clutter,target rotation,with the help of Robust characteristics of Convolutional Neural Network（ CNN） for light,rotation,background clutter and other complex changes emerged in tracking target,the paper proposes the CKCF robust vision tracking algorithm based on Convolutional Neural Network. Considering data security and privacy protection technology,CKCF algorithm uses the picture data set to train VGG model for target feature deep extraction,which could be integrated into the improved KCF tracking algorithm. The experimental results show that compared with the KCF algorithm,the algorithm has more robust tracking effect,and solves the problem of target tracking drift or failure for nuclear related filter KCF algorithm under the complicated conditions such as the illumination changes,background clutter,target rotation.