针对红外目标相关滤波跟踪过程中由于背景杂波干扰、目标遮挡和目标形变等情况导致的鲁棒性差甚至跟踪目标丢失的问题,提出一种融合跟踪-学习-检测方法和相关滤波理论的红外目标跟踪算法.该算法在传统相关滤波框架基础上,融合目标的方向梯度直方图特征和亮度直方图特征,改善了目标轻微形变导致的模型漂移问题.针对背景杂波和遮挡导致的多峰值响应问题,对目标背景区域的相关响应进行惩罚,建立目标和背景响应的多模态检测机制,实现目标由粗到精的定位,并采用自适应的学习率优化跟踪模型的漂移问题；针对目标被严重遮挡或脱离视野的问题,通过全局目标再检测,实现目标的重捕.实验结果表明,在复杂红外地面环境下,该算法有效地解决了相似目标干扰和目标被严重遮挡导致的目标丢失问题.基于OTB-2015视频基准序列和红外视频序列测试,对比多个主流的相关滤波跟踪算法,该算法在跟踪精度和成功率方面较长时相关滤波跟踪算法分别提升了5.6%和4.1%；在目标遮挡指标测试中,该算法在跟踪精度和成功率方面相较长时相关滤波跟踪算法分别提升了4.6%和6.1%.

Aiming at the problem of tracking failure and less robustness caused by background clutter, occlusion and object deformation in infrared object tracking, an infrared object tracking method combining tracking-learning-detection method and correlation filtering theory was proposed. Based on the traditional correlation filtering framework, the proposed method combines the direction gradient histogram feature and the luminance histogram feature to improve the model drift caused by slight deformation of the target. Aiming at the multi-peak response problem caused by background clutter and occlusion, the response of the target background area was punished, and the multi-modal detection mechanism of target and background response was established to achieve the target from coarse to fine positioning, and the adaptive learning rate was used to optimize the drift problem of the tracking model; Aiming at the problem that the object was severely occluded or the object was out of view, the global re-detection of the target was implemented to achieve the target re-capture. The experimental results show that the proposed algorithm effectively solves the object loss caused by background clutter and occlusion in the complex infrared ground environment. Based on the benchmark OTB-2015 and infrared video sequence test, compared with the mainstream correlation filtering tracking algorithms, the proposed algorithm improves the tracking accuracy and success rate by 5.6% and 4.1% respectively compared with the Long-term Correlation Tracking (LCT) algorithm; In the occlusion index test, the proposed algorithm improves the tracking accuracy and success rate by 4.6% and 6.1% respectively compared with the LCT algorithm.