摘要
针对空间正则相关滤波(SRDCF)算法正则权重固定和模型退化等问题,提出了一种基于显著感知与一致性约束的目标跟踪算法。提取方向梯度直方图特征、浅层特征及中层特征进行融合,提升物体外观模型的表达能力;通过显著性检测算法获得初始帧的显著感知参考权重,建立正则权重在相邻2帧之间的关联;最小化实际一致性响应与理想一致性响应之间的差异,约束一致性水平防止滤波器模板退化;提出一种动态约束策略,进一步提高跟踪器在复杂场景下的适应性。在OTB2015、TempleColor128和UAV20L公开数据集上对所提算法进行测试,实验结果表明:相比于SRDCF算法,所提算法在OTB2015数据集上距离精度提高了0.108,AUC提高了0.077,速度为22.41帧/s,实时性较好。
Aimed at the problems that the spatially regularized discriminative correlation filtering(SRDCF)algorithm with fixed regularization weight and model degradation,a correlation filtering tracking algorithm based on saliency awareness and consistency constraint was proposed.Firstly,the histogram of the oriented gradient feature,shallow feature,and the middle feature was extracted to improve the expression ability of the appearance model.Secondly,the regularization weight between the two adjacent frames was associated after the saliency detection algorithm determined the saliency-awareness reference weight of the initial frame.Furthermore,to prevent the degradation of the filter model,the difference between the practical and the scheduled ideal consistency map was minimized and the consistency level was constrained.In addition,a dynamic constraint strategy was proposed to further improve the adaptability of the tracker in complex scenarios.The algorithm is tested on the public OTB2015,TempleColor128,and UAV20L benchmarks.Experimental results show that compared with SRDCF,the proposed algorithm improves the accuracy by 0.108 and the success rate by 0.077 on OTB2015,with a speed of 22.41 frames per second,and has a good real-time effect.
作者
国强
吴天昊
徐伟
CHORNOGOR Leonid
GUO Qiang;WU Tianhao;XU Wei;CHORNOGOR Leonid(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Key Labotatory of Advanced Marine Communication and Information Technology,Ministry of Industry and Information Technology,Harbin 150001,China;Schoolof Radio Physics,Kharkiv V.N.Karazin National University,Kharkiv 61166,Ukraine)
出处
《北京航空航天大学学报》
EI
CAS
CSCD
北大核心
2023年第9期2244-2257,共14页
Journal of Beijing University of Aeronautics and Astronautics
基金
国家重点研发计划(2018YFE0206500)
国家自然科学基金(62071140)
国家国际科技合作专项(2015DFR10220)。
关键词
机器视觉
目标跟踪
相关滤波
显著感知
一致性约束
machine vision
target tracking
correlation filtering
saliency awareness
consistency constraint
