Real time diagnosis of 3d rail profile faults with computer vision using laser cameras

Railway transportation in our country and in the world is a very popular transportation method. The physical structures of the rail tracks, the weight and speed of the rail transport, and the intensity of use of the railroad are all the reasons for various failures such as wear, breakage, cracking and deterioration in the rail track. In order to identify these failures, the quick and efficient maintenance of the periodicals will prevent possible financial and moral losses. In this thesis, methods have been proposed for real-time detection of failures on rail tracks on railways using contactless laser cameras. The laser cameras used to detect and diagnose millimetric level errors on rail tracks first obtain a three-dimensional model of the rail, and then diagnose faults that are actually modeled. In the development phase of the method, methods using feature selection algorithms such as basic component analysis, singular value decomposition, kernel basic component analysis, histogram comparison, and random forest and convolutional artificial neural network machine learning algorithms have been developed on input data for digital camera and laser camera operations. In order to verify the proposed methods, the experimental set established in the thesis was used and validation studies were carried out with real data. As a result, the performance of the methods developed for the diagnosis of failures in railway lines using laser cameras in this doctoral thesis is demonstrated by experimental data and field studies. The studies carried out during the thesis were supported with the research project of TUBİTAK 1001 numbered 114E202.