一种抗遮挡的背景感知相关滤波器

An Anti-Occlusion Background-Aware Correlation Filter

在目标跟踪的应用中,背景相似的情况下的相关滤波器算法具有较好的鲁棒性,但是容易受到遮挡等因素干扰,导致跟踪失败。为了有效跟踪目标,本文提出了一种抗遮挡的背景感知相关滤波器,该滤波器采用多特征融合、尺度跟踪和引入置信度机制来提高跟踪效果的鲁棒性。在OTB100数据集上测试表明,本文算法在尺度位置变化、相似背景、低分辨率、遮挡等情况下都表现出了较高的准确性,尤其在抗遮挡性中表现出了良好的鲁棒性。

In the application of target tracking, the correlation filter algorithm with similar background has good robustness, but it is easily interfered by occlusion and other factors, which leads to tracking failure. In order to effectively track the target, this paper proposes an anti-occlusion background perception correlation filter, which uses multi-feature fusion, scale tracking and confidence mechanism to improve the robustness of tracking effect. The test on OTB100 data set shows that the proposed algorithm shows high accuracy in the case of scale position change, similar background, low resolution, occlusion, etc., especially in anti-occlusion.