Image-based Train Component Integrity Detection Method And System Development

Train maintenance plays an important role in guarding people's safety.The traditional maintenance method mainly relies on the subjective judgment of the inspectors,and the professional technical level of the maintenance personnel is relatively high,and the visual fatigue caused by the long-term operation will inevitably affect the correctness of the subjective judgment,thus affecting the safe travel of the train.At present,train maintenance mainly relies on the combination of manual and computer vision assistance.However,the computer vision technology provides low intelligence in train maintenance,and the image collection area of the detected train components is limited in a greater extent.The advantages of them cannot be well reflected.Based on the existing fault detection system,this thesis designs a method of the train component integrity detection based on image processing.The main research contents of this thesis are as follows.1)The thesis combine with the characteristics of the train components,and the integrity detection idea for the train component image is designed through the demand analysis of the train component data provided by the railway department.2)To solve the problem of complex shape train components and shooting interference,this thesis designs an improved SIFT feature point detection algorithm,which can reduce the number of feature points in the flat area of train components and increase the number of feature points in the sharp position of train components.Therefore,the improved method reduce the number of mismatched pairs and improve the accuracy of feature point matching.The original RANSAC algorithm is improved for the purification of matching pairs,and the time of matching pair purification is optimized.3)Two kinds of the train component integrity detection algorithms are designed based on the registered train components in the thesis.The skeleton of the train components is extracted from the common faults of known train components,and the improved DWT matching is used to identify the integrity of train components.For the faults of the common shape of unknown train components,the abnormal region is located through the analysis of the differential image after registration,and then the contour similarity of the abnormal region is calculated to judge the integrity of the train component.The detection system of train component integrity is tested by using the images of 437 train parts collected in this thesis.The experimental results show that the rate of correct recognition is 88.1%,which can achieve the purpose of assisting manual operation and greatly improve the recognition accuracy of the train component integrity test.