LiDAR Aided Train Track Occupancy Identification Method

Railway is currently the mainstay and one of the main modes of transportation in China's comprehensive transportation systems with the advantages of large capability,low energy consumption and low displacement,which also plays a significant role in the process of economic and social development.In order to ensure the safety of train operation and improve railway transportation efficiency,train localization systems are required to provide the track occupancy state along with the position and the velocity of the train for train control systems.Train autonomous localization based on on-board equipment is the development trend of the next generation of train control systems.Common track occupancy identification based on train autonomous localization is achieved using GNSS measurements,the error level of which may result in wrong identification of track occupancy under certain railway environments.Turnout is the major scenario of track occupancy identification.LiDAR sensors is capable of detecting turnouts,which provide additional track topology so that track occupancy identification can be improved at both parallel track sections and turnout sections.From the standpoint of track occupancy identification within train autonomous localization,the potential problems of the application of track and turnout detection based on LiDAR sensor in track occupancy detection are analyzed.The performance requirements for the LiDAR sensor are proposed on the basis of the geometrical topology of the track.Then a track and detection approach with the interference of environmental noise is presented.Based on Bayesian modeling a train position state estimation framework is established,and within it the GNSS and velocity measurements are fused with the aid of a digital track map to obtain the one-dimensional position of the train.Models of track and turnout detections are built and integrated into the train position state estimation framework to form a track occupancy state identification method based on probabilistic models of track events.Finally an integrated verification of the proposed method is performed using the statistical results which are obtained by processing the field test data.The research contents as below are contained in this dissertation:Firstly,the performance requirements for the LiDAR sensor are presented in view of rail track and turnout detection.Taking into account the spatial distribution characteristics of tracks under different train operation conditions and the appearance in the point clouds of LiDAR sensor,the corresponding performance parameters of the LiDAR sensor are proposed with the constraint of driving speed and specific requirements for track detection.Then a LiDAR calibration method for track detection is developed according to specific setup condition.The features of railway scene and track structure are summarized.The basic scheme of track detection using LiDAR is presented with the reference of object detection methods in road environment.Secondly,a rail track topology detection method based on spatial-temporal features is proposed.According to the rail features within a single scan,the key points of rails are extracted based on model matching.Then track detection is achieved by using the longitudinal continuity of rails and the lateral association feature of tracks.A branching direction pre-detection method based on track topology events is proposed considering the spatial-temporal features of turnout components.This method reduces the distance consumption required by branching direction detection after passing turnouts.Thirdly,a train position state estimation method based on Bayesian modeling is proposed.The pedal curve of the GNSS measurement is obtained using the measurement information.A map matching algorithm which involves direction-dependent GNSS standard deviation is designed,and thus the one-dimensional position of the train is obtained by fusing with velocity measurements in combination with a digital track map.In addition,a train position state estimation framework based on Bayesian modeling is built,which allows the integration of the position variable into the calculation of the probabilities of different track hypotheses.By combining train position determination with track occupancy identification,the method solves the problem of ambiguous determination of train positions at turnout sections and improves the efficiency of track occupancy identification.Lastly,a track occupancy state identification method based on probabilistic models of track events is proposed.Based on the digital track map models are built for turnout distance detections,turnout passing direction detections and track topology detections respectively.The implicit branching hypotheses are extended in order to obtain route based hypotheses which are more applicable for train track occupancy identification scenarios.The accuracy of each kind of detection is analyzed statistically and thus the track events are introduced into the existing Bayesian localization framework.This method addresses the integration of track and turnout detections into the existing train position state estimation framework,which improves the efficiency and accuracy of track occupancy identification remarkably.A software verification system for train track occupancy identification based on a GNSS reveiver,a velocity sensor and a LiDAR sensor was designed.This dissertation used a large number of track occupancy identification scenarios provided by the field test data and the virtual data obtained by flipping and verified that the proposed method can resolve track occupancy identification more efficiently and accurately.The achievements can offer theoretical references to the research and design of train track occupancy identification scheme for the next generation of train control systems and the research on other related critical technologies and technical standards.