隧道火灾的视觉跟踪算法  

Visual tracking algorithm for tunnel fire

作　　者：马庆禄[1,2] 王伟 孙枭 邹政 罗昊 MA Qinglu;WANG Wei;SUN Xiao;ZOU Zheng;LUO Hao(Chongqing Engineering Laboratory for Transportation Engineering Application Robot,Chongqing Jiaotong University,Chongqing 400074,China;School of Traffic and Transportation,Chongqing Jiaotong University,Chongqing 400074,China;The Key Laboratory of Road and Traffic Engineering of Ministry of Education,Tongji University,Shanghai 201804,China)

机构地区：[1]重庆交通大学交通工程应用机器人重庆市工程实验室,重庆400074 [2]重庆交通大学交通运输学院,重庆400074 [3]同济大学道路与交通工程教育部重点实验室,上海201804

出　　处：《东南大学学报(自然科学版)》2025年第1期255-265,共11页Journal of Southeast University:Natural Science Edition

基　　金：国家自然科学基金资助项目(52072054);交通部三峡库区奉建高速公路安全智能建造科技示范工程资助项目(Z29210003);重庆交通大学研究生科研创新资助项目(CYS240482,2024S0075);重庆交通大学交通工程应用机器人重庆市工程实验室2022年度开放课题计划资助项目(CELTEAR-KFKT-202301)。

摘　　要：为了实时掌握隧道现场火灾的发展状况,在YOLOv8算法火灾检测的基础上结合改进后的DeepSORT算法,提出一种火焰和烟雾的视觉跟踪算法YOLOv8-T。该算法使用EfficientNetV2替换原YOLOv8特征提取网络实现对算法的轻量化调整;引入三重注意力(TA)模块和第4个检测头,以提高算法检测精度以及对小目标的检测能力;同时采用ShuffleNetV2替换DeepSORT算法中的ReID模块,并引入DIOU方法代替传统的IOU,在保证跟踪准确度的条件下减少模型计算复杂度,以提升火灾跟踪的实时性。实验结果表明,在隧道火灾跟踪方面,YOLOv8-T算法比SORT算法、DeepSORT算法、YOLOv8+DeepSORT算法在跟踪准确度上分别提高了26.20%、15.86%和9.21%,在跟踪精度上分别提高了11.28%、9.06%和2.2%。在ID变换次数上分别减少22.2、15.3和10.4次,表明YOLOv8-T算法具有较高的火灾跟踪能力。研究成果可为公路隧道火灾监测提供参考,并为实现隧道火灾救援提供依据。To grasp the development status of tunnel fire in real time,combined with the improved DeepSORT algorithm,a visual fire and smoke tracking algorithm,YOLOv8-T,was proposed based on YOLOv8 fire de-tection.EfficientNetV2 was used to replace the original YOLOv8 feature extraction network to achieve the lightweight adjustment of the algorithm.Triplet attention(TA)module and the fourth detection head were in-troduced to improve the detection accuracy and the ability of the algorithm to detect small targets.Meanwhile,ShuffleNetV2 was used to replace the ReID module in DeepSORT,and the distance intersection over union(DIOU)method was introduced to replace the traditional IOU,so as to reduce the computational complexity of the model while ensuring the tracking accuracy,so as to improve the real-time performance of fire tracking.The experimental results show that the tracking accuracy of YOLOv8-T algorithm is improved by 26.2%,15.86%,and 9.21%compared with SORT,DeepSORT,and YOLOv8+DeepSORT algorithms,respec-tively,in tunnel fire tracking.The tracking accuracy is improved by 11.28%,9.06%,and 2.2%,respec-tively.The number of ID changes is reduced by 22.2,15.3,and 10.4 times,respectively,indicating that the YOLOv8-T algorithm has high fire tracking ability.The research results can provide reference for highway tun-nel fire monitoring and tunnel fire rescue.

关 键 词：隧道工程 隧道火灾 跟踪算法 应急救援 

分 类 号：U458.1[建筑科学—桥梁与隧道工程]