抗遮挡与尺度自适应的改进KCF跟踪算法

Improved KCF tracking algorithm based on anti-occlusion and scale transformation

核相关滤波(kernel correlation filter,KCF)目标跟踪算法在目标发生尺度变化或受长时间遮挡时无法实现准确跟踪.对此,提出一种融合窗口滤波的抗遮挡尺度自适应的目标跟踪算法.首先,利用快速尺度估计对跟踪目标进行尺度变换;然后,当检测到目标受到遮挡时停止对分类器模型的更新,融合加权窗口滤波器对目标位置进行预测,并利用预测信息修正目标跟踪区域;最后,将所提出算法移植到DJI Guidance视觉平台上并为其设计跟踪实验.实验结果表明,该算法能够有效解决目标尺度变化与目标受遮挡等跟踪问题.

The kernel correlation filter(KCF)target tracking algorithm can not be effectively applied when the scale of the target changes or the target is occluded for a long time.Therefore,this paper proposes an improved KCF tracking algorithm based on anti-occlusion with fused window filtering and scale transformation.Firstly,the tracking target scale transformation is used to estimate the scale of target fastly.Then,when the target is occluded,the update of the classifier model is stopped,then the weighted window filter is used to predict the target position for the target tracking area.Finally,the algorithm is transplanted to DJI Guidance vision platform,and the experimental results show it can effectively solve the problem of target scale transformation and target occlusion.