面向红外测温的双波段图像融合人脸检测方法  

Dual-band image fusion face detection for infrared temperature measurement

作　　者：李涵钰 李希才 LI Han-yu;LI Xi-cai(School of Electronic Science and Engineering,Nanjing University,Nanjing 210000,China)

机构地区：[1]南京大学电子科学与工程学院,江苏南京210000

出　　处：《激光与红外》2025年第3期444-451,共8页Laser & Infrared

基　　金：国家科技部重点研发计划项目(No.2022YFB3606600);中央高校基本科研业务费专项资金项目(No.2024300443)资助。

摘　　要：为了提高红外热像仪人脸检测准确度,本文提出了一种基于双波段图像融合的人脸检测方法,该方法将可将光(RGB)图像和红外(IR)图像经线性融合后,通过Yolo-FastestV2轻量级卷积神经网络进行人脸检测。相较于传统的红外测温系统中需要分别对红外和可见光图像分别进行人脸检测,本文提出的双波段人脸检测方法仅需要进行一次检测即可同时获得IR图像和RGB图像中的人脸位置,并减少了传统方法在坐标映射阶段由于距离变化而引入的映射误差。为了完成对双波段融合图像的训练和测试,设计了包含可见光和红外的双波段图像数据集,数据集采用双波段相机进展拍摄,该摄像机由可见光传感器和红外探测器组成,可同时采集RGB图像和IR图像。实验结果表明,利用双波段融合的方法能够正确检出测试集中94.35%人脸图像,最高检测帧率可达317 FPS。In order to improve the accuracy of face detection of infrared thermal imaging cameras,a face detection method based on dual-band image fusion is proposed,which can perform face detection through Yolo-FastestV2 lightweight convolutional neural network after linear fusion of visible light(RGB)images and infrared(IR)images.Compared with the traditional infrared temperature measurement system that requires separate face detection for infrared and visible images,the dual-band face detection method proposed in this paper requires only one detection to obtain the face position in both IR and RGB images,and reduces the mapping error introduced by the traditional method due to the distance change in the mapping process in the coordinate mapping stage.In order to complete the training and testing of dual-band fusion images,a dual-band image dataset containing visible light and infrared,and the dataset is shot by a dual-band camera,which consists of a visible light detector and an infrared detector,and the two sensors can simultaneously capture RGB images and IR images.The experimental results show that 94.35%of the face images in the test set can be correctly detected using dual-band fusion,and the highest detection frame rate can reach 317 FPS.

关 键 词：双波融合 红外人脸检测 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN291[自动化与计算机技术—计算机科学与技术]