The Research On Rail Scratch Detection Algorithms

ABSTRACT:Traditional rail abrasion detection mainly depend on artificial work, which is not only high-cost and low-efficient but also that each worker needs to take charge of a long rail line. The probability of leak-detecting will be quite large in this condition. Therefore, an auto method that replaces the traditional way by using machine is urgently needed by railway departments and research institutes.The steel rail abrasion detection algorithm based on computer vision digital image processing techniques not only solve the leak-detecting caused by worker’s visual fatigue, but also greatly improved the efficiency. Generally, a computer can undertake thousands of detecting tasks, which saves the cost markedly. The issues this paper solve can be mainly divided into two classes:corrugation and block abrasion.For corrugation detection, based on the rail image collected by vehicle track inspection system and the morphological feature of corrugation, we present a detection method based on local frequency feature analysis and corrugation images. First, extract the precise track area images through the track-locating algorithm based on weighted position. Then, analysis and extract every feature vector generated by the Fourier transformation energy characteristics of the rail image. And, adopt the SVM(support vector machine) to classify the feature vectors, so that judge whether each column is a corrugation line. Finally, find the continuous corrugation interval, and determine whether this trail image has corrugation according to that.For abrasion detection, this paper mainly combined with the traditional video frame of foreground detection and background modeling algorithm. Divide the track image into several section, then each section corresponds to one video frame, and the whole track corresponds to some short video frame sequences. This paper mainly refers to the idea of ViBe algorithm yet adopts to different features, using more reasonable method. Our experiment show that the abrasion detection algorithm based on local background modeling, by which we obtain the good detecting results, is superior to traditional method, and it can be expanded to other defect-detecting applications.