Environmental Perception Method Research Of Urban Rail Transit Based On Instance Segmentation

The continuous development of urban rail transit has put forward higher requirements for safe operation.Research on the environmental perception of urban rail transit has become a top priority.At the same time,the successful application of environmental perception in the automotive intelligent assisted driving system provides new ideas and solutions to the problem of obstacle detection in urban rail transit.This paper has conducted in-depth research on environment perception and obstacle detection in front of the track.The instance segmentation network is introduced into the obstacle detection problem.The environment perception method based on the Mask R-CNN instance segmentation network is selected to solve the problem of rail traffic obstacle detection.The main work and achievements of this paper include:(1)The importance of studying urban rail transit environment perception is elaborated,and the achievements and problems of domestic and foreign scholars on rail transit environment perception and obstacle detection problems are summarized.On this basis,the design scheme of urban rail transit environment perception based on instance segmentation network is proposed,including an overview of the overall design scheme,the design of visual information collection based on vehicle camera sensors and vehicle host,the design of data set construction,and the instance segmentation network design.(2)In order to fully train the instance segmentation network to achieve a good detection effect,the visual information collection based on urban rail transit is realized through the on-board camera sensor and the on-board host.This solves the serious problem that the current urban rail transit operation scenario has no public data set.Through the collected video information,after screening and calibration,18,000 frames of samples are selected to complete the construction of the operating scene data set.(3)Three instance segmentation algorithms with important influence are introduced.After training and comparison on urban rail transit scene data sets,the Mask R-CNN instance segmentation network with the best comprehensive performance is selected to complete image segmentation,which is used to solve the obstacle detection research task.(4)The backbone convolutional network for feature extraction of Mask R-CNN instance segmentation network is improved.Through comparison of four different backbone convolutional networks,Res Ne Xt-101 is selected to realize the urban rail environment perception based on the Mask R-CNN instance segmentation network.The Mask R-CNN instance segmentation network has completed the accurate image segmentation of the current track,adjacent tracks and obstacles.The detection rate of train obstacles reached 99.9%,and the accuracy rate of image segmentation was as high as 98.4%,which completed the research task of obstacle detection in this paper.