一种鲁棒的基于集成学习的核相关红外目标跟踪算法

在红外目标跟踪中，由于目标所处的背景信息复杂多变和目标外观的显著变化，单一的分类器不足以拟合多模态的数据。该文结合核相关滤波器(KCF)将多个核相关分类器通过集成学习整合到一个框架中。利用KCF分类器具有解析解的特点平衡跟踪鲁棒性与实时性之间的矛盾，从而解决单个分类器无法处理复杂背景与显著的外观变化问题，并显著提升目标跟踪的性能与稳定性。为了验证算法的有效性，该文利用两个核相关跟踪器联合学习出1个强分类器。大量的定性定量实验表明所提的算法的跟踪性能超过传统的KCF算法，且跟踪速度也超过大多数比较算法。

A Robust Kernelized Correlation Tracking Algorithm for Infrared Targets Based on Ensemble Learning

In the infrared object tracking, the single classifier is not enough to fit the multimodal data due to the complex background information of the target and the significant change in the appearance. In this paper, Kernelized Correlation Filters (KCF) tracking algorithm is used to integrate kernelized correlation classifiers into one framework through ensemble learning. It uses the KCF classifier that has analytical solutions to balance the contradiction between the robustness and instantaneity, thereby addressing the complex background and significant appearance changes, and consequently significantly improving the tracking performance and stability. To verify the effectiveness of the algorithm, this paper uses two kernelized correlation trackers to learn a strong classifier. The qualitative and quantitative experiments show that the proposed algorithm outperforms the traditional KCF algorithm, and the tracking speed is superior to most of the comparison algorithms.