针对目标跟踪中因目标遮挡、目标出视野导致的跟踪失败问题, 为提高相关滤波目标跟踪算法的鲁棒性, 提出了一种基于双步相关滤波的目标跟踪算法。首先根据方向梯度直方图特征不同单元大小目标表征的特点, 提出双步相关滤波目标跟踪框架, 在提高目标跟踪精度的同时保证了跟踪速度; 然后融合多种目标特征, 来获得目标更加全面的特征表征, 以提高目标跟踪的鲁棒性; 最后提出基于目标跟踪置信度指标的目标模板自适应更新策略, 来解决目标遮挡时目标模板被污染的问题。实验在OTB100标准目标跟踪数据集上进行验证, 通过与其他跟踪算法进行比较结果表明, 该算法与其中最优跟踪算法相比, 目标跟踪精度提升6.0%, 目标跟踪成功率提升5.5%, 平均跟踪速度为27.4 fps, 保证了目标跟踪的实时性。实际目标跟踪应用中, 在目标严重遮挡等情况下, 该算法仍然可以对目标进行稳定精确地跟踪。

Aiming at the problem of tracking failure caused by object occlusion and out of view in object tracking, a two-step correlation filter for object tracking algorithm was proposed to advance the robustness of the object tracking via correlation filter. Firstly, according to the characteristics of different cell size of histogram of oriented gradient (HOG) feature, a two-step correlation filter object tracking framework was presented, which can improve the tracking accuracy and simultaneously ensure the tracking speed. Then, fusing multiple object features and obtaining more characteristic representation comprehensively were to promote the robustness of object tracking. Finally, an object template adaptive updating strategy was proposed based on the object tracking confidence index, which was to solve the problem that the object template was contaminated when the target was occluded. The experiment was validated on the OTB100 standard object tracking dataset. The results of comparison with other tracking algorithms show that the tracking accuracy is improved by 6.0% and the tracking success rate is increased by 5.5% in comparison with the optimal tracking algorithm, and the average tracking speed is 27.4 fps, which ensures the real-time performance of object tracking. In the application of actual object tracking, the algorithm can still track the target stably and accurately in the case of severe occlusion.