基于核相关的尺度自适应视觉跟踪

A Scale Adapted Tracking Algorithm Based on Kernelized Correlation

针对视觉跟踪中目标尺度变化对准确跟踪的不利影响,提出一种基于核相关的尺度自适应视觉跟踪算法。首先,通过建立核岭回归模型构建二维核相关定位滤波器,采用融合后的多通道特征对滤波器进行训练,提高目标定位的精度;然后,对目标区域进行多尺度采样,样本缩放后提取其特征,并构造为一维特征,以此构建一维核相关尺度滤波器,估计出目标的最佳尺度。在OTB2013平台上的实验结果表明,与8种当前主流的跟踪算法相比,本文算法的跟踪精度和成功率均有优势。在尺度变化条件下,本文算法在快速准确跟踪的同时,较好地实现了对目标尺度的自适应跟踪。

In order to solve the problem of accurate tracking and scale estimation in videos where targets change their scales,we propose a scale adapted tracking algorithm based on kernelized correlation.Firstly,we establish kernel ridge regression model and construct a two-dimensional kernelized correlation location filter.The center location of target is determined precisely by using fused multi-channel features.Then,the multi-scale samples of target area are obtained and their sizes are reset to the same with the model.By extracting their features and reconstructing to one-dimensional vector,we construct the one-dimensional kernelized scale filter to achieve optimal scale estimation.The experimental results on OTB2013 platform,especially on the scale changing benchmark dataset indicate that the proposed algorithm performs better in precision and success rate in comparison with eight mainstream tracking algorithms.Meanwhile,this algorithm can not only achieve an adapted tracking to the scale changing of target,but also locate its position fast and effectively.