尺度自适应在线鲁棒目标跟踪

针对在线boosting跟踪算法在目标外观发生大幅度变化以及遮挡时易产生“漂移”导致目标丢失问题进行了研究，提出一种尺度自适应在线鲁棒目标跟踪算法。算法基于目标灰度或彩色直方图统计特征构建权重图像，通过对权重图像的矩特征分析，可以实现对目标尺度的自适应调整，同时该算法引入半监督学习策略，很好地解决了由于在线学习导致的跟踪失败问题。实验结果表明，本文算法很好地解决了遮挡、目标外观和尺度变化时的鲁棒跟踪问题。与EM-shift，MIL和SPT三种算法相比，跟踪成功率以及鲁棒性均有所提高。

On-line Robust Object Tracking with Scale Adaption

To solve the shift problem that on-line boosting based tracking algorithm often faced due to substantial change of appearance and occlusion, a scale adaptive robust object tracking algorithm is proposed in this work. The algorithm achieves scale adaptive by analyzing the moments of the weight image that based on statistical features of gray scale or color histogram features. Then a semi-supervised strategy is introduced to solve the tracking failure during on-line updating. Video tests show that the proposed on-line robust tracking algorithm achieves robust tracking under the situation of occlusion, appearance change and scale variation. Compared with EM-shift, MIL and SPT algorithm, the proposed method exhibits higher accuracy and robustness.