引用本文:王殿伟,杨旭,韩鹏飞,等.复杂背景下全景视频运动小目标检测算法[J].控制与决策,2021,36(1):249-256
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览次   下载 本文二维码信息
码上扫一扫!
分享到: 微信 更多
复杂背景下全景视频运动小目标检测算法
王殿伟1, 杨旭1, 韩鹏飞2, 刘颖1, 谢永军3, 宋海军3
(1. 西安邮电大学通信与信息工程学院,西安710121;2. 西湖大学人工智能研究与创新中心,杭州310024;3. 中国科学院西安光学精密机械研究所,西安710119)
摘要:
为解决复杂背景下全景视频中运动小目标检测精度低的问题,提出一种基于复杂背景下全景视频运动小目标检测算法.首先,为降低复杂背景信息的干扰,提高目标检测的精度,采用快速鲁棒性主成分分析(Fast RPCA)算法将全景视频图像的前景背景信息分离,并提取出前景信息作为有效的图像特征;然后,改进更快的基于区域的卷积神经网络(Faster R-CNN)中的区域生成网络(RPN)的候选框尺度大小,使之适应全景图像中的目标尺寸,再对前景特征图进行训练;最后,通过RPN网络和Fast R-CNN网络共享卷积层输出检测模型,实现对全景视频图像中小目标的精准检测.实验结果表明,所提出算法可以有效抑制复杂的背景信息对目标检测精度的影响,并对全景视频图像中的运动小目标具有较高的检测精度.
关键词:  全景图像  Fast RPCA  Faster R-CNN  目标检测
DOI:10.13195/j.kzyjc.2019.0686
分类号:TP183
基金项目:陕西省自然科学基金基础研究计划科技创新“双导师制”项目(2018JM6118);西安邮电大学创新创业基金项目(2018SC-08);西安邮电大学研究生创新基金项目(CXJJLY2018038).
Panoramic video motion small target detection algorithm in complex background
WANG Dian-wei1,YANG Xu1,HAN Peng-fei2,LIU Ying1,XIE Yong-jun3,SONG Hai-jun3
(1. School of Telecommunication and Information Engineering,Xián University of Posts and Telecommunications,Xián710121,China;2. Center for AI Research and Innovation,Westlake University,Hangzhou310024,China;3. Xián Institute of Optics and Precision Mechanics of CAS,Xián710119,China)
Abstract:
In order to solve the problem of low detection accuracy of moving small targets in the panoramic video in complex background, a small target detection algorithm based on complex background motion is proposed. Firstly, to reduce the interference of complex background information and improve the accuracy of target detection, the fast robust principal component analysis(Fast RPCA) algorithm is used to separate the foreground background information of the panoramic video image, and the foreground information is extracted as an effective image feature. Then, the candidate frame size of the region proposal network(RPN) in the faster region-convolutional neural networks(Faster R-CNN) is improved to adapt to the target size in the panoramic image, and then the foreground feature map is trained. Finally, the convolutional layer output detection model is shared by the RPN network and the Fast R-CNN network to achieve accurate detection of small targets in the panoramic video image. Experiments show that the proposed algorithm can effectively suppress the influence of complex background information on target detection accuracy, and has high detection accuracy for small moving targets in panoramic video images.
Key words:  panoramic image  Fast RPCA  Faster R-CNN  target detection

用微信扫一扫

用微信扫一扫