自适应更新上下文相关滤波补充学习跟踪算法

为了使跟踪算法在遇到背景混乱、遮挡等问题时,能具有更好的鲁棒性,提出了一种自适应更新的上下文感知相关滤波补充学习跟踪算法(adaptive context-aware correlation tracking,ACCT)。首先,利用上下文感知相关滤波计算出响应图;接着,在更新阶段,加入了峰旁比(peak-to-sidelobe,PSR)置信度检测机制,自适应地更新相关滤波器模型;最后,与颜色直方图分类器进行线性融合,得到最终的响应值,响应值最大的位置即跟踪结果。实验上,将所提方法与先进的实时跟踪算法进行对比,结果表明,所提算法在跟踪标准测试集OTB50和TC128上,相比基准算法成功率有显著提高;并且能很好地解决背景混乱、目标被遮挡等问题。

Real-time Tracking Complementary Learners Tracking With Adaptive Update Context Correlation Filter

In order to make the tracking algorithm more robust, when object suffer from background clutters and occlusions. In this paper, an adaptive context-aware correlation tracking algorithm(ACCT) is proposed. Firstly, the response map is calculated by context-aware correlation filter. Then, in the update step, the reliability detection mechanism of Peak-to-Sidelobe (PSR) is added, and the correlation filter model is adaptive updated. Finally, the final response value is generated through combining the color response and the context-aware correlation filter response. The position of maximum response is the final tracking result. Experimentally, the results show that the proposed tracker favourable performance against several state-of-the-art tracker on the OTB50 and TC128 datasets and, it can well solve the problems such as background clutters, object occlusion.