摘要
由于非制冷红外热像仪采集的红外图像具有对比度低、纹理模糊的特点,仅仅利用灰度特征很难实现稳定的追踪性能,严重制约复杂环境下武器装备精确打击的能力。在多视角协同学习的基础上,利用脊回归具有解析解的优势,通过最小化类内方差来分析样本之间的类内结构关系,提出一种基于最优间隔分布与多视角特征协同的红外目标跟踪算法,可以同时优化类内间隔方差和类间间隔方差,实现红外目标的实时精确追踪;针对传统模型参数更新适应性不强的问题,引入峰值旁瓣比对跟踪参数进行自适应更新,提升模型的泛化能力。大量定性定量仿真实验结果表明,提出的红外目标追踪算法在目标发生部分遮挡、运动模糊、背景干扰、旋转以及灰度变化等复杂环境下,能够较准确地追踪目标,具有重要的理论和应用研究价值,适合高性能低成本非制冷红外导引头。
Due to the low contrast,unclear texture in the infrared image obtained by the low-cost infrared detector,it is difficult to achieve stable tracking performance,which severely restrict the precise attack capability of infrared missile seekers.On the basis of multi-view collaborative learning,this paper uses the advantage of analytic solution of ridge regression to minimize the intra-class structure relationship between samples by minimizing the intra-class variance,and proposes an object tracking algorithm based on optimal interval distribution and multi-view features learning,which can simultaneously optimize the intra-class interval variance and the inter-class interval variance so as to achieve accurate tracking for infrared objects in real time.Aiming at the problem of poor adaptability of traditional model parameters updating,an adaptive updating strategy based on peak sidelobe ratio is used to improve generalization ability.Simulation results show that the improved algorithm has obvious advantages in real-time,stability and quantitative indexes,and is suitable for high-performance,low-cost,miniaturized uncooled infrared seekers.
作者
赵蔷
谢鹏
ZHAO Qiang;XIE Peng(School of Computer Science,Xianyang Normal University,Xianyang 712000,P.R.China;School of Equipment Management and UAV Engineering,Air Force Engineering University,Xi’an 710000,P.R.China)
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2020年第4期639-647,共9页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
陕西省教育厅专项科研计划项目(15JK1803)
陕西省教育科学“十三五”规划2017年课题(SGH17H197)
咸阳师范学院教育教学改革研究资助(2015Z004)
陕西省2018年大学生创新创业训练计划项目(201828036)。
关键词
目标追踪
红外图像
多特征融合
协同学习
峰值旁瓣比
object tracking
infrared image
multi-feature fusion
collaborative learning
peak sidelobe ratio
