基于时间一致性约束的核互相关器目标跟踪

针对相关滤波器跟踪算法在目标快速运动、遮挡和表观变化时易发生跟踪漂移或者丢失的问题,提出一种基于时间一致性和核互相关器的目标跟踪算法。该算法通过引入对图像噪声和杂波更具鲁棒性的核互相关向量,能够更精确地预测目标的仿射变化。同时,在学习过程中引入时间一致性约束,以解决因核相关器时间退化导致的跟踪漂移问题。最后,采用主灰度分量逆映射来提升跟踪器应对目标部分遮挡的能力。在公开的OTB100标准目标跟踪数据集中与提供的基准算法和其他性能更加先进的相关滤波算法进行对比,该算法平均跟踪速度为41f/s,相对fDSST和SAMF算法,其跟踪精度分别提升15.6%和6.4%,跟踪成功率分别提升33.3%和6.1%。实验结果表明,该算法在目标快速运动、遮挡或表观变化时仍能精确地跟踪目标。

Object Tracking with Kernel Cross-correlator Based on Temporal Consistent Constraint

For the problems of tracking drift or loss in the tracking algorithms of correlation filter, which is caused by fast motion, occlusion and appearance variation of objects, an object tracking algorithm with kernel cross-correlator based on temporal consistent constraint is proposed. A kernel cross-correlator vector, which is more robust to image noise and clutters, is introduced to predict affine transformation of object more precisely. Meanwhile, to solve the problem of tracking drift caused by temporal degradation of kernel cross-correlator, temporal consistent constraint is introduced during learning process. Finally, MGC(major gray component) anti-projection is utilized to improve the ability of the tracker to deal with occlusion of object. The proposed algorithm is compared with provided benchmark algorithms and other more advanced correlation filter based algorithms on public OTB100 standard object tracking datasets. The tracking speed of proposed algorithm precision reaches 41f/s, and compared with fDSST and SAMF algorithm, its tracking precision is increased by 15.6% and 6.4%, and the success rate is increased by 33.3% and 6.1% , respectively. The experimental results demonstrate that the proposed algorithm is able to track objects precisely under the condition of fast motion, occlusion and appearance variation.