基于多尺度特征融合的自主巡检视频识别摄像头图像配准方法  

Camera Image Registration Method for Autonomous Patrol Video Recognition Based on Multiscale Feature Fusion

作　　者：胡延国 牛亚东 李子龙 张程亮 李游 Hu Yanguo;Niu Yadong;Li Zilong;Zhang Chengliang;Li You(Everbright Environmental Protection(China)Co.,Ltd.,Shenzhen 518000,Guangzhou,China;Everbright Environmental Energy(Boluo)Co.,Ltd.,Huizhou 516139,Guangzhou,China;China Everbright Environment Group Limited,Shenzhen 518000,Guangzhou,China)

机构地区：[1]中国光大环境(集团)有限公司,广东深圳518000 [2]光大环保能源(博罗)有限公司,广东惠州516139 [3]光大环保(中国)有限公司,广东深圳518000

出　　处：《科技通报》2024年第5期51-54,共4页Bulletin of Science and Technology

摘　　要：针对自主巡检视频识别摄像头图像受结构相似性影响导致配准精度低的问题,本文提出基于多尺度特征融合的自主巡检视频识别摄像头图像配准方法。在图像增强的基础上,采用Harris角点检测方法获取自主巡检视频识别摄像头图像结构特征,并基于特征向量构建结果,对图像特征进行多尺度融合。设计图像配准变换估计模型实现自主巡检视频识别摄像头图像配准。结果表明:在特征匹配时,本文所提方法可以有效缩短匹配时间使得融合结果更为精确,具有良好配准效果。Aiming at the problem of low registration accuracy caused by the structural similarity of autonomous patrol video recognition camera images,a method for autonomous patrol video recognition camera image registration based on multi-scale feature fusion is proposed.On the basis of image enhancement,Harris corner detection method is used to obtain the structural features of autonomous patrol video recognition camera images,and based on the feature vector construction results,a multi-scale fusion of image features is performed.Design an image registration transformation estimation model to achieve autonomous patrol video recognition camera image registration.The results show that the proposed method can effectively shorten the matching time and make the fusion results more accurate in feature matching,with good registration results.

关 键 词：HARRIS角点检测 特征向量 图像识别 特征匹配 图像配准 

分 类 号：TN249[电子电信—物理电子学]