面向室内环境控制的人员信息检测系统的设计与实现  

Design and Implementation of a Personnel Information Detection System for Indoor Environment Control

作　　者：张文利 郭向 杨堃 王佳琪 朱清宇 ZHANG Wenli;GUO Xiang;YANG Kun;WANG Jiaqi;ZHU Qingyu(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;China Construction Technology Group Co.,Ltd.,Beijing 100070,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]中建科技集团有限公司,北京100070

出　　处：《北京工业大学学报》2020年第5期456-465,共10页Journal of Beijing University of Technology

基　　金：“十三五”国家重点研发计划资助项目(2016YFC0700306-06)。

摘　　要：为了自动获取室内环境中人员信息(数量、性别、体表温度信息),实现对室内环境设备进行有效调节及控制,提升室内人员的舒适度,提出一套基于红外热图像及可见光图像融合的面向室内环境控制的人员信息检测系统.系统通过同步采集室内场景中的可见光图像及红外热图像,并利用相机的视场角进行图像配准.针对传统基于面部的人员检测方法易受到侧脸、背身等头部多姿态的影响,而导致检测精度下降的问题,采用基于更快速区域卷积神经网络(faster region-based convolutional neural network,Faster R-CNN)的头部检测算法在可见光图像中精准检测人员的头部位置,并进行人数统计;依据头部区域采用基于深度学习的性别检测算法,判断人员性别;将可见光图像中捕获的人员头部区域映射在红外热图像对应位置,利用红外热图像的温度标尺以及颜色映射关系,自动计算得到室内人员的体表温度,提高系统的独立性和可操作性.实验结果表明,本系统可以自动准确地检测实际场景中的室内人员,并获取人员个数、性别以及体表温度,实现对上述信息的有效统计和可视化显示,为控制室内环境提供良好的技术基础和数据支撑.To automatically obtain information(number,gender,and body temperature)of indoor people effectively,adjust and control indoor equipment,and improve the comfort of indoor personnel,an information detection system of people for indoor environment control based on infrared thermal image and visible image fusion was proposed in this paper.First,visible light images and infrared thermal images in the indoor scene was acquired synchronously,and images were matched by useing camera’s field of view angle.To tackle the problem that the traditional face-based person detection method was easily affected by facial occlusion and head multi-pose,which caused the decline in detection accuracy,the faster region-based convolutional neural network(Faster R-CNN)head detection algorithm was used to accurately detect and count people in visible light images.According to the head region,a gender detection algorithm based on deep learning was used to determine the gender of the person.Further more,human head regions captured in the visible light image were mapped to corresponding positions of infrared thermal images.Using the temperature scale of infrared thermal image and the color mapping,the body temperature of the indoor personnel was automatically calculated to improve the independence and operability of the system.The experimental results show that the system can automatically detect people and obtain the number,gender and body temperature of the people.At the same time,the above mentioned information can be effectively statistically and visually displayed,which provides a technical basis and data support for controlling the indoor environment.

关 键 词：人员信息检测 红外热图像 图像融合 深度学习 更快速区域卷积神经网络(Faster R-CNN) 室内环境 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]