Research On High Security Environmental Perception Technology Of Train Based On Basic Point Slope Method

The latest train visual perception technology combines camera and lidar as vision equipment.The image captured by the camera is determined by deep learning to determine the position of the track.The point cloud data collected by lidar are clustered to get the position of the object in the field of vision.The combination of the two can detect whether foreign objects appear on the track.This technique relies entirely on the image to locate the track.When the image information is invalid,the lidar can not determine whether the object is in the track,and the train will lose its detection function.In order to supplement this defect,the point cloud data collected by lidar are analyzed in this paper,proposed the basic point slope method to extract the track boundary,so that the system can identify the position of track by the data collected by lidar,and improve the stability of perception system.The research proves that it is feasible for lidar to identify orbit separately,it can improve the stability and security of train vision system based on multi-sensor fusion.The main contents of this paper are as follows:(1)This paper classifies and collates the technology of track line detection,and finds that there is no precedent for the application of single sensor of lidar in track environment sensing.The existing train environment sensing system does not use lidar for track recognition.(2)The data processing method of vehicle-borne lidar for road detection is investigated in detail,and the difference of structural characteristics of railway from road are analyzed.How to combine road detection method with railway features to realize the function of track environment perception is studied.(3)By using the elevation classification method,the horizontal projection of the point cloud is meshed,and the maximum elevation of the points in the grid is classified to judge and divide the location of the track.(4)Based on the idea of smoothing feature,the point cloud is layered according to the distance.The base slope method is proposed to extract the track boundary points.According to the characteristics of track,the boundary points are screened and the track boundary lines are fitted by the least square method.(5)The deviation degree between the fitting track and the actual track position is counted,and the possible sources of the error are pointed out.Then it is verified that the point cloud can detect the track when the image information is invalid,which reduces the missing rate of the environmental perception system.It is proved that the idea of using lidar for track recognition to improve the train operation security is feasible.The paper includes 31 figures,4 tables and 33 references.