Robust Visual Tracking With Spatial Regularization Kernelized Correlation Filter Constrained by a Learning Spatial Reliability Map
Robust Visual Tracking With Spatial Regularization Kernelized Correlation Filter Constrained by a Learning Spatial Reliability Map
复制标题
DOI：
10.1109/access.2019.2902216
复制
发表时间：
2019
期刊：
影响因子：
3.9
通讯作者：
Qianbo Liu;Guoqing Hu;Md Mojahidul Islam
中科院分区：
文献类型：
--
作者：
Qianbo Liu;Guoqing Hu;Md Mojahidul Islam
视觉跟踪作为计算机视觉中的一个基础研究课题，由于跟踪问题的复杂性，如突然运动、视线外、变形、严重遮挡等，仍然具有挑战性。在本文中，我们通过引入空间正则化分量来惩罚核化相关滤波器系数，从而扩展了核化相关滤波器(CF)以实现稳健跟踪。为了提供更可靠的预测，我们构建了一个基于颜色直方图的空间可靠性图，以加强目标中心附近的检测样本。进一步利用特征融合和模型更新机制来提高跟踪的有效性。在OTB-2013、OTB-2015和Temple Color-128数据集上进行了广泛的实验。综合结果表明，与这些数据集上具有代表性的航迹相比，本文提出的方法具有一定的优越性。
As a basic research topic in computer vision, visual tracking is still challenging because of the complexity of the tracking problems, such as abrupt motion, out-of-view, deformation, and heavy occlusion. In this paper, we extend the kernelized correlation filter (CF) for robust tracking by introducing spatial regularization components to penalize the CF coefficients. To afford a more confident prediction, we construct a spatial reliability map based on the color histogram to enforce the detecting samples near the target center. The feature fusion and the model update mechanism are further employed to improve the effectiveness of tracking. The extensive experiments are executed on the OTB-2013, OTB-2015, and Temple Color-128 datasets. The comprehensive results demonstrate the superiority of our proposed method comparing to the representative tracks on these datasets.