Research On Train Obstacle Detection And Recognition Technology Based On Deep Learning

The rapid development of urban rail transit not only improves the urban traffic environment,facilitates the communication between regions,but also promotes the economic construction of the city.At present,the automatic driverless metro train has been put into operation in urban rail transit.In this technology,the detection of obstacles is the key technology to realize the development of the automatic driverless metro.Based on the deep learning theory,this paper designs and implements an obstacle detection method which is suitable for the active environment perception of driverless metro train.Through the data fusion of vision and lidar sensor,the obstacle detection in front of metro train is finally completed.The main work of this paper is as follows:First of all,this paper uses the instance segmentation algorithm to complete the identification and segmentation of the rail region.This instance segmentation algorithm extracts the features of the rail area through CNN(Convolutional Neural Networks).On this basis,the segmentation algorithm is applied to process the extracted features to complete the accurate segmentation and recognition of the rail area.In addition,this paper uses the real urban rail train operation environment data to make the rail data set for network training.Secondly,this paper realizes obstacle detection by multi-sensor data fusion technology.Based on the ROS(Robot Operating System),the plane calibration method is applied to calibrate the vision sensor;then,through the joint calibration of the vision sensor and the lidar sensor,the mapping from the lidar sensor to the vision sensor is realized,and the multi-sensor data fusion is completed;After the data fusion,the target recognition algorithm is applied to detect the obstacles in the running track area of the train,and the early warning is carried out when the obstacles are detected.Finally,according to the method and the characteristics of subway train,an experimental platform for obstacle detection is designed.The experimental platform includes hardware and software,the hardware part is the design of the main control board in the system host;in order to achieve the overall function of the system,based on the ROS system,the upper computer software module for data fusion and obstacle detection is compiled.Through the experimental verification,the results show that the proposed method has high robustness,which will be of great significance to improve the safety of train operation and transportation efficiency.