Locomotive Number Automatic Recognition System Based On Image Processing

With the accelerate development of China’s High-Speed Rail, routine inspection andmaintenance of locomotives have become the essential recently. Locomotive number recognition isthe one of a key technology in locomotive online detection system. In this paper, based on the existinglocomotive number recognition, corresponding improved methods are put forward in positioning,segmentation and recognition of locomotive number by using image processing and the system oflocomotive number automatic recognition has been accomplished.Considering of the inaccurate positioning on locomotive number, segmentation obtained fromearly positioning is adopted to guide final positioning. With the prior knowledge of the size, locationand arrangement of the locomotive number, the proposed method can judge the result of positioningaccording to some features, such as the size of region, the size of characters, and its arrangementmode. The experiments have shown that the method can obtain the accurate positioning for differenttypes of images of locomotive number.It is very difficult for global thresholding to obtain the correct segmentation of locomotivenumber. For this situation, the edge information is applied for guiding the thresholding. Different fromthe conventional segmentation method, several segmentation results of the same locomotive numberobtained by different ways are judged based on the size of locomotive number and its ratio of thelength and width of locomotive number according to the prior knowledge of preparation rules andpainting standards of the locomotive number of different locomotives. The experiment has shown thatthe proposed method can segment the locomotive number which has error results by conventionalsegmentation and then improve the accuracy and practicality.Locomotive number uses the straight character, including particular characters of multiple forms,which may cause breaking and adhesion sometimes, and this problem affects the accuracy oflocomotive number identification. To solve this problem, a valid eigenvector-based identificationmethod is put forward. Firstly, an improved template matching method is adopted, to select charactersamples of skeleton features for template match, and the preliminary matches are get with the featurevalue, which is the ratio of the pixels of sample skeleton features projected on the template and theskeleton itself. Secondly, with the samples auxiliary features of connected domain value, stringingvalue, network lattice vector and the character location, an effective character feature vector iscomposed, which is used to correct the preliminary match results so as to get correct recognition results. The experiment indicates that this method have better adaptability in the situations ofcharacter breaking, tilt, rotation, deformation and etc. And the different expression of the samecharacter can be quickly told so that we can get the correct recognition results.Finally, the locomotive automatic train identification system is designed and realized in the paper.The effectiveness and practicality have been verified in the final trail.