摘要
针对复杂环境下的有轨电车障碍物入侵检测问题,提出了基于实例分割的障碍物入侵检测及轨道识别方法。首先建立有轨电车运行环境数据集;再基于Yolact实例分割模型对数据集的数据进行训练、验证和测试,得到障碍物检测和轨道区域提取模型;最后将待测图像输入模型后输出障碍物轮廓、类别、轨道区域轮廓数据,进而提出障碍物与轨道的相对位置关系判断方法、障碍物是否入侵轨道判断方法。实验表明,该方法模型能以像素级的精度准确识别复杂场景下的障碍物和轨道,并准确判断障碍物在轨道的位置,可作为有轨电车运行环境综合感知的重要部分,提高运行安全。
Aiming at the obstacle intrusion detection of trams in complex environments,we proposed an obstacle intrusion detection and track recognition method based on instance segmentation.Firstly,a data set of tram operating environment was established.Based on the Yolact model,the data set was trained,verified and tested.Obstacle detection and track area extraction models were obtained.Finally,the images to be tested were inputted into the model to obtain the obstacle contour,category,and track area contour data.Furthermore,a method for judging the relative position between obstacles and the track,and a method for judging whether the obstacle invaded the track was proposed.Experiments showed that the method model could accurately identify obstacles and tracks in complex scenes with pixel-level accuracy,and accurately determine the location of obstacles on the track.The recognition method had high accuracy and was able to be used as an important part of the comprehensive perception of the tram operating environment to improve operational safety.
作者
黄世泽
杨玲玉
陶婷
宋冠群
HUANG Shize;YANG Lingyu;TAO Ting;SONG Guanqun
出处
《上海公路》
2021年第2期89-94,103,M0006,共8页
Shanghai Highways
基金
上海市科委重点项目(19DZ1209402)。
关键词
有轨电车
障碍物入侵检测
轨道识别
实例分割
tram
obstacle intrusion detection
track recognition
instance segmentation
