Semantic Segmentation And Extraction Of 3D Laser Point Cloud Orbit Objects Based On Deep Learning

Railway is not only one of the national infrastructure,but also one of the necessary elements to promote the rise of the national economy as a means of transportation for the whole people.The survey project of the existing railway line is one of the necessary steps for the maintenance and repair of the railway and the addition of the second line.The traditional method of manual static measurement with railway central line as control line has some problems,such as low safety factor and limited operation status of railway,which results in low efficiency of the method.The mobile measurement technology,which integrates three-dimensional laser scanning,panoramic camera and inertial navigation positioning and attitude sensor,is a new technology gradually developed in the field of Surveying and mapping in the past 20 years.This technology utilizes navigation technology based on GNSS,INS,DMI and IMU to obtain positioning and attitude information of laser scanning mobile platform,and it combines three-dimensional laser scanner and high-definition camera equipped on the platform to obtain observation data with geographic reference information in mapping scenes.Compared with the traditional manual static mapping method,the threedimensional laser scanning mobile mapping system based on mobile measurement technology has incomparable advantages in data acquisition efficiency,data richness and data acquisition security.In terms of measurement accuracy,the accuracy of mobile measurement also keeps pace with the development of GNSS,INS,camera,laser scanner and other hardware and integrated navigation algorithm.Although the three-dimensional laser scanning mobile mapping system solves some problems encountered in the stage of data acquisition of three-dimensional laser point clouds in railway scenes,there are still many difficulties in the classification and extraction of point clouds.The main manifestation is that the existing three-dimensional laser point cloud segmentation and extraction methods are lack of application in railway scenes,and the traditional visual interpretation method of extracting rail,turnout,railway signal lights and other characteristic information is inefficient,and inevitably made mistakes in interpretation.Therefore,a method of semantics segmentation and extraction of 3D laser point cloud orbital objects based on depth learning is proposed in this paper.The purpose of this method is to apply deep learning algorithm to semantics segmentation and recognition of railway track objects,and to extract track lines on this basis.In order to implement this method,the main research contents of this paper are divided into three parts:(1)Research on the orbital object segmentation and classification technology based on deep learning.By understanding the existing deep learning algorithms,the paper studies the method of using them to segment and classify rail objects in railway scenes.Then,according to the research results,the main body of the deep learning algorithm is improved,so that it can deal with the problem of three-dimensional laser point cloud processing in railway scenes more pertinently.(2)Research on the orbital line extraction technology based on three-dimensional laser point cloud.Aiming at the three-dimensional laser point cloud data of rails,the projection algorithm of rail point cloud and the accurate matching algorithm between the twodimensional cross section of rail point cloud and the standardized structure information of rails are studied.Then the coordinates of railway track centers are calculated according to the matching results.Finally,the complete three-dimensional railway track line is formed by connecting,and the technology of track line extraction based on three-dimensional laser point cloud is realized.(3)Verification of the semantics segmentation and extraction method and accuracy evaluation.That is to say,the method of semantics segmentation and extraction of railway point clouds is experimented,and the accuracy of the experimental results is evaluated.According to the evaluation results,the experimental conclusion of semantics segmentation and extraction of railway point clouds is drawn.