Research On The Technology And Method For Automatic Operation Of On-Board DMI Of Train Control System

The current on-board equipment functional test of train control system has the characteristics of high repeatability and long test cycle,which consumes a lot of human resources.Therefore,effectively improving the automation and intelligence level of onboard equipment functional testing has become a key issue to be solved urgently.Driver-machine Interface(DMI)serves as a window for feedback of test results,as well as a window for instruction delivery and data entry.It is an effective means to improve the automation and intelligence of the test platform to realize automatic information identification,command judgment and key operation of DMI.Based on machine vision technology,this thesis designs the on-board DMI automatic operation platform of the train control system from the perspective of modularization.The platform is analyzed and verified from both theoretical and practical aspects.The main research work of this thesis is as follows:(1)Summarizes the categories of information to be identified on the DMI interface.According to the functional test requirements of on-board equipment,the DMI global information is divided and extracted in detail,16 types of key information and related characteristics that need to be identified in the test process are determined.According to the feature difference of the information,a method for text icon class information recognition based on deep learning target detection algorithm is proposed,and a reading recognition scheme of pointer speed instrument panel based on traditional image processing method is designed.(2)The recognition of DMI text icon information is completed by using the deep learning target detection algorithm.By comparing the performance of different deep learning target detection algorithms in the DMI data set,the icon text information recognition scheme with the YOLOv4 algorithm as the core is determined.Aiming at the problem that the data set is too single,the data set expansion is completed based on the Mosaic data augmentation method.By improving the K-means clustering algorithm,the accuracy of the anchor fitting data set is improved by 25.46%.For the small target detection problem,the loss function calculation method is improved based on the focal loss.The detection performance of the improved YOLOv4 algorithm on small targets has been significantly improved.On the DMI dataset,the m AP value is improved by2.6%,and the recognition frame rate reaches 11.21 frames per second.(3)The reading recognition of the pointer speed instrument panel is realized using the traditional image processing method.In order to highlight feature information,median filtering,linear change of image gray value and threshold segmentation based on OTSU algorithm are used in turn to complete DMI dashboard images preprocessing.The equation of the pointer line was fitted by RANSAC least squares method.The center coordinates of the instrument panel are determined by weighting fusing the coordinates of the center of the circle fitted by the center of gravity of the tick marks and the coordinates of the intersection of the line fitted with the tick marks.Finally,the speed is determined by the offset angle of the pointer relative to the zero scale line.The average single image recognition time is 0.31 s,and the recognition accuracy rate can reach 99.85%.(4)Designed and implemented the on-board DMI automatic operation platform.The platform architecture is designed based on the idea of modularization.The on-board DMI automatic operation platform is built by the combination of software development and physical processing,and a joint test is carried out with the on-board equipment function test platform.Experiments show that the system runs stably and efficiently.Figure 75,Table 25,77 References.