摘要
针对背景杂乱、遮挡、热交叉以及目标形变等复杂跟踪场景下目标跟踪算法出现性能严重退化问题,提出一种基于自适应标签和稀疏学习相关滤波的实时红外单目标跟踪算法。首先,根据目标响应情况自适应地构造样本标签,通过自适应标签训练提升相关滤波器的分类能力,抑制干扰区域对跟踪模型的污染。其次,加入稀疏学习策略,通过目标响应L1范数抑制复杂跟踪场景下目标响应多峰分布,提高跟踪算法的鲁棒性;与基线算法相比,该算法精度和AUC分别提升了19.3%和39.8%。在数据集GTOT、RGBT234和VOT-2016TIR上的实验结果表明,该算法对上述复杂跟踪场景具有良好的应对能力,运行速度超过35 fps,综合性能优于对比跟踪算法。
To deal with the serious performance degradation of target tracking algorithms in complex tracking scenes, such as background clutter, occlusion, thermal crossover, and target deformation, a real-time infrared single object tracking algorithm based on the adaptive label and sparse-learning correlation filter is proposed. First, sample labels are constructed based on the target response adaptively, and the discrimination ability of the correlation filter is enhanced by training with adaptive sample labels, which suppresses the pollution of the interference region to the tracking model. Secondly, the sparse learning strategy is introduced to suppress the multi-peak distribution of the response map in complex tracking scenes by its L1 norm, resulting in improving the robustness of the tracking algorithm. Compared with the baseline algorithm, the precision and AUC of the proposed algorithm are improved by 19.3% and 39.8%, respectively. Experimental results on datasets GTOT, RGBT234, and VOT-2016TIR show that the proposed algorithm has a strong ability to deal with the above complex tracking scenes. Its running speed is over 35 fps, and its comprehensive performance is better than the compared tracking algorithms.
作者
黄月平
李小锋
卢瑞涛
齐乃新
张胜修
Huang Yueping;Li Xiaofeng;Lu Ruitao;Qi Naixin;Zhang Shengxiu(College of Missile Engineering,Rocket Force University of Engineering,Xi′an 710025,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2022年第12期199-208,共10页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(62006240)
陕西省自然科学基础研究计划(2021JQ-373)项目资助
关键词
计算机视觉
红外目标跟踪
相关滤波
稀疏学习
computer vision
infrared object tracking
correlation filter
sparse learning
