相比传统目标跟踪算法,基于相关滤波的目标跟踪算法跟踪精度更高、实时性更好,但当目标被遮挡或超出视野时,难以提取目标特征、正确定位检测窗口,容易导致目标跟踪失败或漂移。因此,提出了一种基于置信度的并行相关滤波跟踪算法。首先,使用一种新的置信度评价方法判断目标是否被遮挡或存在异常情况。其次,在置信度的基础上,用组合权重融合两种不同的跟踪器,构建并行相关滤波跟踪算法,以提高算法的跟踪精度和鲁棒性。最后,为了防止模型污染,对两种滤波器模型采取自适应权重更新策略。在OTB-2013和OTB-2015数据集上进行实验,结果表明,相比传统算法,本算法在跟踪精度和成功率上均有显著提升。

Compared with traditional target tracking algorithms, the target tracking algorithm based on correlation filter has great advantages in tracking accuracy and real-time performance. However, when the target is blocked or out of view, it is difficult to extract target features and correctly locate the detection window, which may easily lead to target tracking failure or drift. Therefore, in this paper, a parallel correlation filter tracking algorithm based on confidence is proposed. First, the paper proposes a new confidence evaluation method to determine whether the target is blocked or out of view. Second, on the basis of confidence, two different trackers are fused with the combination weight of confidence to construct a parallel correlation filter tracking algorithm in order to improve the tracking accuracy and robustness. Finally, in order to prevent model pollution, an adaptive weight update strategy is adopted for the two filter models. Experiments on OTB-2013 and OTB-2015 datasets show that compared with traditional algorithms, this algorithm has significantly improved tracking accuracy and success rate.