基于激光雷达与视频结合的铁路异物探测技术研究  

A Study on Railway Foreign Object Detection Technology Based on Combination of LIDAR and Video

作　　者：余博 杨森 王珣[1] 裴起帆 潘兆马 YU Bo;YANG Sen;WANG Xun;PEI Qifan;PAN Zhaoma(China Railway Eryuan Engineering Group Co.,Ltd.,Chengdu 610031,China)

机构地区：[1]中铁二院工程集团有限责任公司,成都610031

出　　处：《高速铁路技术》2023年第1期48-53,共6页High Speed Railway Technology

摘　　要：本文提出一种基于激光雷达与视频结合的铁路异物探测系统,该系统利用RANSAC算法和欧氏距离聚类获取轨道平面及异物,基于TBD策略、Kalman和匈牙利算法实现多目标跟踪与状态识别,通过联合标定实现雷达和视频联动并获取清晰的异物图像,最后采用YOLOv4模型识别异物。铁路场景验证结果表明:(1)该系统将能检测到最小目标(投影长度为20 cm)的极限距离延长至75 m,比市场同类系统提升66%;(2)实现了仅由10组数据即可完成的激光雷达与摄像球机的快速标定;(3)采用YOLOv4对图像分类,该系统的m AP指标达到90.2%,大幅度降低误报的可能性。研究成果可在灾害频发地段的铁路运营安全保障中发挥重要作用。This paper proposes a railway foreign object detection system based on the combination of LIDAR and video,in which the track plane and foreign objects are identified with the RANSAC algorithm and Euclidean distance clustering,TBD strategy,Kalman and Hungary algorithms are employed for multi-target tracking and status recognition,LIDAR and video linkage are realized through joint calibration to obtains the clear images of the foreign objects,and finally the YOLOv4model is used to identify the foreign objects.The verification results of railway scenarios show that:(1)The system extends the limit distance for detecting smallest targets(with a projection length of 20 cm)to 75 m,which is 66% higher than that of similar systems in the market.(2)The system realizes the rapid calibration of LIDAR and dome camera that can be completed by only 10 sets of data.(3)By using YOLOv4 to classify images,the mAP index of the system reaches 90.2%,which greatly reduces the possibility of false alarms.This system can play a vital role in the safety assurance of railway operation in disaster-prone areas.

关 键 词：异物入侵探测 激光雷达 TBD策略 联合标定 YOLOv4 

分 类 号：U298[交通运输工程—交通运输规划与管理]