空间相位相关的快速视觉目标跟踪方法

Fast Visual Tracking with Spatial Phase Correlation

在相位相关目标跟踪中,由于响应函数为冲击函数,不能适应目标的变化,且易受噪音和杂波的影响,导致跟踪失败.而均值合成滤波器利用符合期望的相关输出分布,可获得较好的滤波效果.本文考虑跟踪过程中目标的上下文信息,设计与目标空间位置相关的响应函数取代冲击函数,从而获得更为鲁棒的滤波器;利用峰值旁瓣比评估跟踪结果并自适应更新滤波器,适应目标的变化;用快速傅里叶变换进行加速算法,对目标实现快速鲁棒跟踪.实验结果表明,本文方法平均每秒处理50帧图像,在准确度和精确性方面优于当前算法.

The visual tracking based on phase correlation is always failure, because the response is dirac delta function disturbed by the noise and cluster. The desired correlated output distribution is adopted in average of synthetic exact filters, which obtains the excellent filtered result. This paper designs the response related with the object% spatial position to replace the dirac delta function, which can achieve a robust filter. In order to adapt the changed object, peak-to-sidelobe ration was used to evaluate the results to update the filter. The fast Fourier transform was used to accelerate the algorithm to track the object fast and robustly. The experiments show that the proposed method can process 50 frames per second, and excels the state-of-the-art methods in both accuracy and precision.