中文摘要：

提出了一种基于分布场的全局匹配搜索的实时目标跟踪算法,克服了原始分布场的局部搜索和实时性差的局限。采用相关系数代替原始算法的L1范数度量目标分布场与候选区域分布场的距离,有利于运用傅里叶变换,将相关系数从计算复杂度高的时域转换到计算复杂度低的频域来实现,并且能一次算出目标分布场和检测区域所有候选分布场的相似度,从而保证算法的实时性和全局搜索能力,克服稀疏采样方法的随机性和局部结果最优性。实验结果表明,与最近代表性的跟踪算法相比,提出的方法在多个具有挑战性的视频序列中,在平均误差、跟踪速度和成功率上获得了最佳的性能。

英文摘要：

This paper presented a real-time and global search tracking algorithm based on distribution field （DF） to overcome the limitation of local search and poor real-time in original DF tracking. The correlation coefficient instead of L1 -norm was used to measure the distance between target model DF and candidate region DF. With the aid of Fourier transform, correlation coef- ficient computed in the frequency domain with low computational complexity rather than the time domain so that the similarities between target DF and all candidate region DFs would be got fast. It could ensure the algorithm＇ s real-time performance and global searching ability, and then overcame the sparse sampling methods＇ limitation of randomness and local optimality. Com- pared with recent tracking algorithms, experimental results show that the proposed method gets the best performance in the av- erage error and tracking speed in a number of challenging video sequences.