基于结构信息建模和判别稀疏的红外小目标跟踪方法  

Infrared small target tracking method based on structural information modeling and discriminant sparsity

作　　者：木尼拉·塔里甫[1] 安尼瓦尔·加马力[2] 亚森·艾则孜[3] Munila TALIFU;Anniwaer JIAMALI;Yasen AIZEZI(School of Information Management,Xinjiang University of Finance and Economics,Urumqi 830012,China;Experimental Training-Teaching Center,Xinjiang Police College,Xinjiang Police College,Urumqi 830011,China;Department of Information Security Engineering,Xinjiang Police College,Xinjiang Police College,Urumqi 830011,China)

机构地区：[1]新疆财经大学信息管理学院,新疆乌鲁木齐830012 [2]新疆警察学院实验实训教学中心,新疆乌鲁木齐830011 [3]新疆警察学院信息安全工程系,新疆乌鲁木齐830011

出　　处：《光学技术》2021年第5期622-631,共10页Optical Technique

基　　金：国家自然科学基金资助项目(61762086)。

摘　　要：为了提高背景杂波和成像噪声等干扰下红外小目标的跟踪精度,提出了一种基于结构信息建模和判别稀疏的红外小目标跟踪方法。小目标信号在广义高斯目标超完备字典上被稀疏分解,以便从受噪声干扰和杂波污染的红外图像中提取出小目标的空间结构信息;设计了转移受限粒子滤波跟踪算法,以提高粒子的采样概率;在转移受限粒子滤波框架下,基于判别稀疏表示和L1范数最小化框架求解候选目标的稀疏系数,实现小目标的跟踪。基于各种红外序列对所提方法进行实验论证,实验结果表明,所提方法能够在杂波和噪声较大的干扰下稳定地跟踪小目标,其中心误差、重叠率和平均视频播放帧率分别为3pixel、0.7和40fps,均优于其他对比方法,且具有较强的鲁棒性。In order to improve the tracking accuracy of infrared small target under the interference of background clutter and imaging noise,an infrared small target tracking method based on structural information modeling and discriminant sparsity is proposed.Firstly,the small target signal is sparse decomposed in the generalized Gaussian target super complete dictionary to extract the spatial structure information of the small target from the infrared image which is corrupted by noise and clutter.Then,the particle filter algorithm is improved,and the transfer constrained particle filter tracking algorithm is designed to improve the sampling probability of particles.Finally,in the framework of transfer constrained particle filter,the sparse coefficients of candidate targets are calculated based on discriminant sparse representation and L1 norm minimization framework to achieve small target tracking.Experimental results based on various infrared sequences show that the proposed method can track small target stably under the interference of clutter and noise,and its center error,overlap rate and average video playback frame rate are 3 pixel,0.7 and 0.632 fps,respectively,which are better than other comparison methods and have strong robustness.

关 键 词：红外小目标跟踪 结构信息建模 广义高斯目标超完备字典 判别稀疏表示 转移受限粒子滤波框架 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]