摘要
融合红外图像的热源目标和可见光图像的清晰背景可以实现低照度条件下与场景关联的异常行为识别。现有以特征匹配为主的融合方法,受监控场景下可见光与红外成像的尺度、视角、目标特性等差异影响,配准及融合效率及准确性受限。针对该问题,本文提出了基于显著性检测的不同视角下红外与可见光图像融合方法。通过预设热敏感目标,计算可见光与红外的视场转换模型,预先配准红外与可见光视场。使用Mask R-CNN网络提取红外图像中的行人目标显著性区域,根据视场转换模型点将每个目标区域与可见光图像局部融合。最后,通过违规入侵行为辨识为目标进行实验验证。实验表明,论文提出的融合方法能够有效地将红外图像的热敏目标信息与可见光的场景进行融合,可以准确地判断是否发生违规入侵行为。
The fusion of the thermal target of infrared image and the clear background of visible image can realize the recognition of abnormal action associated with the scene under low illumination condition.The existing fusion methods which mainly focus on feature matching are affected by the differences of visible and infrared imaging scales,views,and target characteristics in monitoring scenes,and the efficiency and accuracy of registration and fusion are limited.To solve this problem,this paper proposes a fusion method of infrared and visible images from different views based on saliency detection.By presetting the heat-sensitive target,the transformation model of visible and infrared view is calculated,and the infrared and visible views are registered in advance.The saliency area of pedestrian target in infrared image is extracted with the Mask R-CNN network,and each target area is locally fused with visible image according to the view transformation model point.Finally,the identification of illegal intrusion action is verified by experiments.Experiments show that the fusion method proposed in this paper can effectively fuse the thermal target information of infrared image with the visible scene,and can accurately determine whether the illegal intrusion has occurred.
作者
李一白
王彦林
闫禹
胡敏涛
刘宾
陈平
LI Yi-bai;WANG Yan-lin;YAN Yu;HU Min-tao;LIU Bin;CHEN Ping(College of Information and Communication Engineering,North University of China,Taiyuan 030051,China;Hong Kong-Zhuhai-Macao Bridge Authority,Zhuhai 519060,China)
出处
《激光与红外》
CAS
CSCD
北大核心
2021年第4期465-470,共6页
Laser & Infrared
基金
国家自然科学基金项目(No.61971381)资助。
关键词
图像融合
红外图像
视角变换
显著性检测
image fusion
infrared image
view transformation
saliency detection
