Research On Recognition Technology Of Insulators In Electrified Railway Based On Computer Vision

Detectting and recognizing insulator pictures accurately and efficiently is the premise of bad condition detection of insulator, insuring that all the subsequent testing works perform in the targeted local images, condition detection and detection algorithm get easier, so that detection speed is improved and the detest results become more accurate.In this thesis, efficient and clear images collected by railway integrated inspection cars were taken as samples to realize the efficient recognition of electric railway insulator by some computer vision technologies. Firstly, two main current target recognition methods, template matching and statistical pattern recognition were introduced. And then two template matching ways based on gray and feature were used to do the insulator recognition experiment. At last, with statistical pattern recognition method, three insulator features of Haar, LBP, HOG were extracted for training and then recognize the insulators by using the classifiers get before.For electric railway insulator recognition with template matching methods, three characteristic matching methods based on SIFT, SURF, and ORB were adopted. By recognition and comparison among the three methods, the results showed that SURF’s integrated performance was better than the two others’. In order to comparing the recognition effect of the three matching methods, a friendly MFC interface which was able to realize the whole recognition process perfectly was projected. Matching methods based on the three characteristics of SIFT, SURF and ORB were taken to achieve insulator detections.AdaBoost and SVM training classifiers were used in statistical pattern recognition method. For Haar feature and LBP feature, AdaBoost algorithm to train classifiers was used. By evaluating the generated classifiers, it turned out that altering the intercept ways, size and quantity of the positive and negative samples could improve the performance of the training classifiers obviously. For HOG feature, SVM training classifiers were used. The results showed that changing the accept ways, size, munber of positive and negative samples could also increase the recognition effect in some way.Under the environment of Visual Studio 2013 and OpenCV3.0, experiments realized electrified railway insulator identification by programming. The results showed that for the insularor recognition based on picture matchings, elimilating the incorrect matchings could promote the recognition function apparently. For insulator recognition based on statistical pattern recognition method, the change of the positive and negative samples’s intercept ways, size and amount could make the recognition quite more efficient. Adopting proper accept methods, size and quantity of samples to get a classifier with LBP features, which could perform with high precision and speed.