尺度自适应的压缩跟踪算法

Scale adaptive compressive tracking algorithm

针对原始压缩跟踪使用固定大小的跟踪框来跟踪目标,提出一种尺度自适应的压缩跟踪算法,在原始的压缩跟踪算法的基础上加入粒子滤波方法,利用分类器的响应产生粒子权重,根据粒子权重大小重新采样,从而避免了粒子退化,利用一个2阶的状态转换模型去估计目标的当前位置和尺度大小,使得跟踪算法能适应运动目标的尺度变化。实验结果表明,与原始的压缩跟踪算法相比,该算法在视频流中的跟踪性能得到提升。

The fixed-size tracking box is usually employed to match the target in the traditional compressive tracking algorithm. Much success has been demonstrated in most cases, while it may fail if the target scale changes a lot. In this paper it proposes a scale-adaptive compressive tracking algorithm that can optimize the tracking of the moving target. The algorithm exploits particle filter method on the basis of original compressive tracking algorithm. The classifier response is utilized to generate particle importance weight and a resample procedure preserves samples according to weight to avoid the degeneracy of the particle. Furthermore, a 2-order transition model is presented to estimate the current position and scale status which enable the algorithm to adapt to the scale change of moving targets automatically. The experimental results show that the proposed scale-adaptive compressive tracking algorithm performs favorably against several state-of-the-art tracking algorithms on challenging sequences in terms of the center location and the overlap rate.