摘要
为预防施工现场火灾事故,提高火焰检测性能,设计基于单点多盒检测器(SSD)和MobileNet模型相结合的火焰检测和预警机器人。首先,处理采集的火焰视频图像数据集,并将其划分为真实火焰图片与疑似火焰图片2类;其次,结合迁移学习思想,微调SSD_MobileNet模型中的网络参数值;最后,将获得的检测模型集成到机器人视频监控平台,进行火焰检测与预警,并在某工地上测试其应用效果。结果表明:在检测模型50%的置信度阈值下,机器人能够准确检测出工地场景中的火焰,多帧视频准确率达到90%以上,并可实现实时预警。
In order to prevent fire accidents at construction sites and improve fire detection performance,a detection and early warning robot based on combination of SSD and MobileNet model was designed.Firstly,data set of collected fire video images was processed and divided into two categories,namely real fire images and non-fire images.Secondly,network parameter values in SSDMobileNet model were finetuned based on idea of migration learning.Finally,the obtained model was integrated into robot video surveillance platform for fire detection and early warning,and its application effect was tested on a construction site.The results show that the robot can accurately detect fire in construction sites with a 50%confidence threshold of detection model.Accuracy of multi-frame video exceeds 90%,and real-time warning can be achieved.
作者
李继超
郭聖煜
孔刘林
原毅璨
孙阳
LI Jichao;GUO Shengyu;KONG Liulin;YUAN Yican;SUN Yang(School of Economics and Management,China University of Geosciences,Wuhan Hubei 430074,China;Faculty of Engineering,China University of Geosciences,Wuhan Hubei 430074,China;School of Mechanical Engineering and Electronic Information,China University of Geosciences,Wuhan Hubei 430074,China)
出处
《中国安全科学学报》
CAS
CSCD
北大核心
2021年第4期141-146,共6页
China Safety Science Journal
基金
国家自然科学基金青年基金资助(71801197)
中国地质大学(武汉)中央高校基本科研业务费专项资金资助项目。