Research On Sensor Data Fusion Algorithm Based On Deep Learning In Obstacle Detection Of Trains

With the continuous improvement of modern science and technology,the development of urban rail transit has been valued by more and more cities.The development of drones,driverless cars and unmanned boats has driven the development of unmanned subways for urban rail transit.The detection of obstacles is related to the safe operation of trains,which is the focus and difficulty of research.In this thesis,by studying the methods of current contact obstacle detection and the obstacle detection methods of unmanned vehicles at home and abroad,an obstacle detection method suitable for unmanned subway trains is designed and proposed.This method uses depth learning theory to realize obstacle detection by sensor data fusion.Based on the operating characteristics of subway trains and the obstacle detection methods of unmanned vehicles,designs the overall scheme of this thesis:based on the deep learning theory,a visual sensor and a lidar sensor are selected as the detection tools.A method suitable for the detection of obstacles in the subway train track is proposed.Firstly,the deep learning theory is applied.The position of the rail is detected by the camera,and the dangerous area during the running of the urban rail train is drawn,when there are obstacles in the dangerous area,the system detects them through the method of lidar and video images.The obstacles known to the camera(such as pedestrians)can be jointly detected by the camera and the lidar,which improves the accuracy of the detection;for obstacles unknown to the camera(such as stones of different sizes and shapes),detection of unknown obstacles is achieved by lidar sensors.Finally,the obstacle detection is performed by writing an obstacle detection interface.When no obstacle is detected,the green light is on.When an obstacle is detected,the red light is on,and an emergency brake is generated to ensure the safety of the passenger and the train.Through the study of depth theory and the analysis of existing obstacle detection methods,and according to the braking principle of subway trains,the system structure designed for the detection method of this thesis is designed.The system structure of the method includes a system host,a laser radar and a camera,wherein the system host includes an industrial computer and a main control board.The parts of the main control board constitute the hardware component;the software part includes the lower computer software that drives the relay action and the upper computer software for detecting obstacles in the track.Finally,the non-contact obstacle detection method for subway trains proposed in this thesis is applied to the field test and test based on the "next generation subway car" of the 13th Five-Year Science and Technology Support Project proposed by China CRRC Changke Co.,Ltd.The test results prove that the proposed non-contact obstacle detection method is feasible.