Research On Obstacle Detection Method In Front Of Train Based On Stereo Vision And Radar Data Fusion

Foreign object intrusion detection is a vital project for the safe operation of rail transit.China’s railway covers a wide area and the terrain environment is complex.Natural disasters or random limit violations by people and objects have a significant impact on the safe and stable operation of trains.The existing railway foreign matter monitoring methods mainly include driver visual observation,manual patrol,fixed-point monitoring at frequent accidents or installation of protective nets,which can not realize the real-time monitoring of train operation environment.With the rapid development of intelligent monitoring of rail transit,using automatic equipment to realize foreign matter detection has become a realistic demand that can not be ignored.Through the analysis of the existing railway foreign object detection methods at home and abroad,this thesis adopts the detection method based on vehicle binocular stereo vision and radar data fusion to detect the obstacles in front of the train.The main research contents of this thesis include:(1)The imaging characteristics of the rail in the image are analyzed,and the straight-line curve switchable hybrid model is designed as the fitting model of the rail.The straight rail parameters are gained by using the Progressive Probabilistic Hough Transform in the myopia field area.For the far-field area,the Newton method is used to search the rail boundary points linearly,and the gray characteristics of the rail are considered for verification and screening.The detected rail curve model is expanded in equal proportion to obtain the image safety clearance model.(2)The left and right cameras are calibrated offline,and the captured image sequence in front of the train is stereo matched to obtain the parallax map.The obstacle and rail line plane are disassembled into a combination of different planes,and the identification of foreign objects is completed according to the line segment characteristics of different planes in the UV parallax diagram.The weighted gray Hough transform is used to extract the line segment in the image to obtain the position information of the obstacle,and the coordinates of the object detection frame and the curve of the image safety limit model are used to judge whether it is an intruded object.(3)The error correction experiment is designed.Through the analysis of the measured radar data,the corrected lateral distance error is controlled within 0.15 m.The radar real-time positioning is matched with the electronic track route map to obtain the front route information.Combined with the railway clearance standard,radar parameters and error correction results,the detection area model at the current time is established.By comparing the radar detection target coordinates with the detection area,the obstacle limit invasion results at the current time are obtained.(4)The spatio-temporal fusion model is established,and the multi-sensor fusion strategy based on decision-making level fusion is designed.The radar data frame with slow acquisition time interval is used as a trigger to trigger the sampling operation of the camera,so as to realize the time registration of the two;Through spatial coordinate transformation,the visual detection results and radar detection results at the same time are unified and calibrated in the image.Judge the target consistency of the fused detection results and classify the risk level.Tested in different scenarios,the results show that the application scenarios of multisensor fusion detection using stereo vision and radar are richer,which overcomes the disadvantage of limited detection of single sensor and effectively reduces the missed detection rate.The average detection accuracy of this method is 89.51%.The research results have important practical significance and practical value for ensuring the traffic safety of rail transit.