基于视觉注意力的Gabor滤波器跟踪方法  

Gabor Filter Tracking Method Based on Visual-attention

作　　者：王慧敏 孙建武 WANG Huimin;SUN Jianwu(Army Academy of Armored Forces,Beijing 100072,China)

机构地区：[1]陆军装甲兵学院,北京100072

出　　处：《装甲兵学报》2022年第6期92-100,共9页Journal of Armored Forces

摘　　要：针对战场环境中多装甲目标间的遮挡、穿插以及目标尺度不断变化等问题,提出了一种基于视觉注意力的Gabor滤波器在线多装甲目标跟踪方法。通过模拟视网膜结构构造了一种能够增强检测的视觉注意力Gabor滤波器分支,并引入时间信息,采用在线学习的目标特性卷积神经网络解决目标遮挡的问题,进而采用实际拍摄和互联网下载等手段构建了多装甲目标跟踪数据集,最后对目前成熟的多目标跟踪方法和本文提出方法进行对比。结果表明:本文提出的方法不仅具有优异的跟踪性能,而且能够满足实际的应用需求。To solve the problems of shielding,interpenetration among multi-armored targets and the constant change of target scales in the battlefield environment,an online multi-armored target tracking method based on visual-attention Gabor filter is proposed.A branch of visual-attention Gabor filter that can enhance detection is constructed by simulating the retinal structure,and time information is introduced.The problem of target shielding is solved by using convolutional neural network of target characteristics based on an online learning.A multi-armored target tracking dataset is constructed by means of actual shooting and downloading from the internet.Finally,the current mature multi-target tracking method is compared with the method proposed in this paper through experiment.The results show that the method in this paper not only has excellent tracking performance,but also can meet the actual application requirements.

关 键 词：多装甲目标 在线跟踪 视觉注意力 GABOR滤波器 卷积神经网络 

分 类 号：TJ811[兵器科学与技术—武器系统与运用工程] TP391.41[自动化与计算机技术—计算机应用技术] TN713[自动化与计算机技术—计算机科学与技术]