Research On Method Of Railroad Track Obstacle Detection Based On Multi-sensor Fusion

Railroad transportation network is an important part of China’s transportation infrastructure,and the safety of rail transportation is of great significance to cross-city transportation.The intelligent detection of rail obstacles can eliminate the safety hazards caused by iron obstacles,which can ensure the safety of railroad operation and people’s travel safety.For the current problem of low accuracy of the detection technology of railroad track traffic obstacles,after analyzing the characteristics of train travel,this paper studies the detection method of track obstacles based on multi-source sensors by using visual sensors,radar sensors and other acquisition equipment.The multi-source data fusion scheme and obstacle detection depth network for the environment around the railroad and the detection of obstacles in the environment are designed,so as to improve the accuracy and reliability of obstacle detection.The specific research contents are as follows:(1)Environment and obstacle detection method around railway tracks in visual environment.Based on the images of railroad tracks and their environment captured by the camera in front of the vehicle head of the train,we design each branch of area segmentation,curve segmentation and target detection applicable to the railroad track environment by analyzing the visual features of the targets in the railroad track environment,and realize the extraction of the features of railroad track area,railway track line and obstacles,and realize the segmentation of railroad track area,railroad track line and railroad track detection based on YOLOP multitasking network for the special scenes of railroad The detection of obstacles.Using the collected visual data,a small training dataset is constructed,and the network is tested and analyzed based on real train operation environment data.(2)Multi-sensor fusion with rail surroundings and obstacle detection method.To address the problem of low accuracy of rail track obstacle detection by single sensor,we design multisource sensor time synchronization scheme and space-time fusion scheme based on PTP protocol and Autoware calibration tool respectively by analyzing the spatio-temporal characteristics and features of data collected by vision sensor and Li DAR sensor,and build a rail track obstacle detection method based on Li DAR and vision sensor.For the difference of distance coverage between Li DAR and vision sensors,design the result-level fusion scheme under the two sensors at different distances and scenes,and take advantage of the joint advantages of the two sensors to realize the multi-sensor obstacle detection scheme under different scene conditions.(3)Testing and application of environment and obstacle detection system around railway track.In order to verify the effectiveness and value of the research results of this paper,the fusion detection rate test and analysis for the research results of this paper were carried out using the test site of Hunan China Railway Group.The advantages of the multi-sensor fusion detection method and its value in railroad traffic safety and transportation efficiency are verified through experiments and comparative analysis with the single vision sensor method and the single LIDAR sensor detection method.