尺度自适应结构输出目标跟踪算法

Structured Output Target Tracking Algorithm with Scale Adaptation

针对目标尺度明显变化时采用固定尺度的结构输出目标跟踪算法容易出现跟踪失败的问题,提出一种改进的尺度自适应目标跟踪算法。新算法在传统结构输出跟踪算法基础上,将目标运动信息引入候选样本采集过程,通过自举滤波器的状态转移模型预测目标的当前位置和尺度,生成一组多尺度候选样本集,避免了固定尺度的密集均匀采样,实现尺度自适应的同时降低了算法的计算量。实验结果表明,所提算法在目标发生明显尺度变化、部分遮挡以及旋转等情况下具有较高的鲁棒性,且实时性相比于传统结构输出跟踪算法明显提高。

A new multi-scale tracking algorithm is proposed to solve the problem that the structured output tracking algorithm with fixed scale often leads to failure when the size of the target change obviously. Based on the original structured output tracking algorithm, the proposed algorithm introduces the velocity information of the moving target into the sampling process of candidate samples. A state transition model of the bootstrap filter is used to estimate the current position and scale, generate a set of multi-scale samples and avoid dense sampling with fixed scale, this al-low to realize scale adaptation and reduce the calculation of algorithm. Experiments show that the proposed algorithm has strong robustness when the scale of target changed obviously or target is partially occluded, and achieve higher real-time performance than original structured output tracking algorithm.