摘要
针对消防红外图像分辨率差、对比度低、信噪比低、视觉效果模糊、人体姿态复杂,多障碍物遮蔽、人体姿态不完整等特点。本文提出了一种基于U-Net网络的消防红外图像的人体检测算法,通过该算法解决了消防场景中人体姿态复杂,多障碍物遮蔽,人体形态不完整的困难。同时对比于传统目标检测算法以及YOLO v3算法,本文提出的算法在消防红外图像的人体检测上无论是检测的精度还是运算的实时性上都有大幅的提升。
Aiming at the characteristics of poor resolution,low contrast,low signal-to-noise ratio,blurred visual effect,complex human posture,multi-obstacle occlusion,incomplete human posture and so on of fire infrared images.In this paper,a human body detection algorithm based on U-Net network for fire infrared image is proposed.The algorithm solves the difficulties of complex human body posture,multi-obstacle occlusion and incomplete human body shape in fire scene.At the same time,compared with the traditional target detection algorithm and the YOLO v3 algorithm,the proposed algorithm has greatly improved the detection accuracy and real-time operation of fire infrared image human body detection.
作者
张骏
范彬
杨新军
ZHANG Jun;FAN Bin;YANG Xin-jun(National Special Display Engineering Research Center,State Special Display Engineering Laboratory,Aviation Industry Corp Huadong Photoelectric Company Limited,Wuhu 241002,China;Air Military Representative Office of Army Assembly Department in Shanghai,Shanghai 200000,China)
出处
《激光与红外》
CAS
CSCD
北大核心
2019年第12期1483-1489,共7页
Laser & Infrared
基金
芜湖市科技重点项目基金项目资助
关键词
红外热成像
U-Net
人体检测
检测精度
infrared thermal image
U-Net
human detection
measuring precision
