带目标恢复的无人机长期自适应相关滤波跟踪算法  

Long⁃term adaptive correlation filter tracking algorithm for UAV with target recovery

作　　者：马浩然 黄鹤[1,2] 林国庆[3] 高涛[4] 王会峰[1,2] 张科[1,2] Ma Haoran;Huang He;Lin Guoqing;Gao Tao;Wang Huifeng;Zhang Ke(Key Laboratory of Intelligent Expressway Information Fusion and Control of Chang′an University,Chang′an University,Xi′an,710064,China;School of Electronic and Control Engineering,Chang′an University,Xi′an,710064,China;School of Energy and Electrical Engineering,Chang′an University,Xi′an,710064,China;School of Datascience and artificial intelligence,Chang′an University,Xi′an,710064,China)

机构地区：[1]长安大学西安市智慧高速公路信息融合与控制重点实验室,西安710064 [2]长安大学电子与控制工程学院,西安710064 [3]长安大学能源与电气工程学院,西安710064 [4]长安大学数据科学与人工智能研究院,西安710064

出　　处：《南京大学学报(自然科学版)》2025年第1期105-116,共12页Journal of Nanjing University(Natural Science)

基　　金：国家自然科学基金(52172379);陕西省重点研发计划(2024GX-YBXM-288);中央高校基本科研业务费(300102325501);西安市智慧高速公路信息融合与控制重点实验室(长安大学)开放基金(300102323502)

摘　　要：针对无人机跟踪地面运动目标过程中目标场景发生遮挡、光照变化等复杂情况下,相关滤波模型易退化进而导致跟踪精度降低等问题,提出了一种针对无人机对地面目标的稳定实时跟踪算法.首先,利用FHOG,CN和灰度等特征构建外观融合模型,提高对复杂场景的适应性;然后,在相关滤波跟踪的基础上,设计自适应局部时空正则化策略,在跟踪器中引入空间正则化来实现像素级的滤波器限制,同时设计时间正则化来优化滤波器的更新;其次,对滤波器进行通道可靠性融合,设计了自适应模型更新策略,防止滤波器退化,提高目标定位的精确度;最后,设计目标恢复模块来提高跟踪器强度,能更好地应对复杂环境.实验结果表明,提出的算法与同类文献算法相比,能更好地适应无人机对地面目标复杂场景的跟踪任务并满足实时性.To address the problem that the correlation filtering model is easy to degenerate and leads to the decrease of tracking accuracy in the process of UAV(Unmanned Aerial Vehicle)tracking ground moving targets,such as occlusion and illumination change of the target scene,a stable tracking real⁃time tracking algorithm for UAV to ground targets is proposed.Firstly,the appearance fusion model is constructed by using the features of FHOG,CN and grayscale,which improves the adaptability to complex scenes.Then,based on the correlation filter tracking,an adaptive local spatio⁃temporal regularization strategy is designed.Spatial regularization is introduced into the tracker to achieve pixel⁃level filter constraints,and time regularization is designed to optimize the filter update.Secondly,channel reliability fusion is performed on the filter,and an adaptive model update strategy is designed to prevent filter degradation and improve the accuracy of target positioning.Finally,the target recovery module is designed to improve the strength of the tracker and better cope with the complex environment.The experimental results show that the proposed algorithm can better adapt to the tracking task of UAV to the complex scene of ground targets and meet the real⁃time requirements compared with the similar literature algorithms.

关 键 词：目标跟踪 外观融合模型 正则化 通道可靠性 目标恢复 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]