铁道运输与经济2024，Vol.46Issue(1)：184-191,198,9.DOI:10.16668/j.cnki.issn.1003-1421.2024.01.24
基于改进Mask-R-CNN网络的轨道交通运行环境感知技术应用研究
Application Research of Rail Transit Operating Environment Perception Technology Based on Improved Mask-R-CNN Network
摘要
Abstract
In order to enhance the rail transit train operation environment sensing capability and further ensure the safety of train operation environment,a rail transit operation environment sensing technology scheme based on improved Mask-R-CNN network was proposed for the problem of boundary intrusion of various obstacles in front of train operation.The paper outlined the design requirements for a train operation environment perception system and presents its structural design.The research introduced the Mask-R-CNN deep vision model,and discussed the applicability of its functions to the perception requirements of train operating environment,used the lightweight MobileNetV2 backbone network and PointRend module to improve and optimize the Mask-R-CNN network and further improve the real-time performance and accuracy of the network.Real-world train operation data from the train's forward perspective were collected to create an in-house dataset,and the model was trained using this dataset in combination with publicly available datasets.It has been verified that the F-Measure evaluation value of the improved model reaches 95.56%,demonstrating high accuracy,real-time performance,and ease of deployment.This solution can meet the emerging requirements of the railway traffic industry and represents a new exploration of computer vision technology in the field of railway traffic.关键词
轨道交通/列车运行环境感知/Mask-R-CNN/MobileNetV2/PointRendKey words
Rail Transit/Train Running Environment Perception/Mask-R-CNN/MobileNetV2/PointRend分类
交通工程引用本文复制引用
郑泽熙,范楷,邓晶雪..基于改进Mask-R-CNN网络的轨道交通运行环境感知技术应用研究[J].铁道运输与经济,2024,46(1):184-191,198,9.基金项目
中国国家铁路集团有限公司科技研究开发计划课题(L2021G003) （L2021G003）
中国铁道科学研究院集团有限公司科研项目(2021YJ100) （2021YJ100）
