提出一种基于高置信度模型更新策略的多相关滤波器协同跟踪算法。利用卷积网络结构VGG-Net-19提取目标周围区域的多层卷积特征,构造深度滤波器,以自适应的特征融合策略实现目标初定位;建立尺度滤波器以检测目标的尺寸变化;利用主次峰坡度比作为跟踪置信度指标,设计一种高置信度下的模型更新策略;当跟踪置信度不足时,通过EdgeBox方法提取目标候选区域,利用设计的重检测滤波器,确定目标的最终位置。在标准数据集OTB-100和TC-128上的实验结果表明,本文算法取得了较高的跟踪精度,在目标发生遮挡、光线变化、出视野等复杂情况时,算法依旧可以稳健地跟踪。

A multi-filter collaborative tracking algorithm based on high-confidence updating strategy is proposed. First, the multi-layer convolutional features of the region around the target are extracted using VGG-Net-19, which is a convolutional network architecture, followed by an adaptive feature fusion strategy with the designed deep filter to get the initial position of the target. Meanwhile, a scale filter is constructed to detect the size change of the target. Then, a tracking confidence indicator named primary and secondary peak slope ratio is utilized, which helps to build a high-confidence model updating strategy. Finally, when the confidence is insufficient, the object region proposals are extracted by EdgeBox method, and the final position of the target is determined by the designed re-detection filter. The experimental results on OTB-100 and TC-128 datasets show that the proposed algorithm achieves high tracking precision and also tracks steadily under some complex circumstances, such as occlusion, illumination variation, and out-of-view.