【摘要】 针对目标跟踪算法在目标外观变化等复杂场景下鲁棒性差的问题,提出一种基于孪生网络的双分支目标跟踪算法。首先,通过相似性学习训练判别分支生成目标外观的特征表达;然后,通过图像分类任务训练鲁棒分支生成目标语义的特征表达,同时在鲁棒分支中加入沙漏结构的残差注意力模块进一步编码目标全局信息;最后,对两个独立分支生成的响应图进行加权融合。在跟踪基准数据集OTB2015上进行实验,结果表明,该算法的成功率较基准算法SiamFC提升了8.2%,精度提升了10.2%,同时在光照变化、遮挡、尺度变化等多种复杂场景中均具有良好表现。更多还原

【Abstract】 Aiming at the problem of poor robustness of object tracking algorithm in complex scenarios such as target appearance changes,a two-branch target tracking algorithm based on Siamese network is proposed. First,the discriminative branch is trained through the similarity learning task to generate appearance feature,Then,the robust branch is trained through the image classification task to generate the semantic feature. At the same time,the residual attention module of hourglass structure is added to the robust branch to further encode the global information of the target. Finally,the response map generated by the two independent branches are weighted fusion. Experiments were conducted on the tracking benchmark dataset OTB2015. The experimental results show that compared with the benchmark algorithm SiamFC,the success rate of the proposed algorithm is increased by 8.2%,and the accuracy is increased by 10.2%. At the same time,it demonstrated good performance in various of complex scenarios,such as illumination changes,occlusions,and scale changes.更多还原