Grayscale-information-based Segmentation Registration for Fault Diagnosis of Train Components

Considering the long time consumption and high labor price in manual maintenance of train systems, computer-vision-based maintenance schemes have become more and more popular, which can detect the abnormality of train components efficiently by virtue of train images. Nevertheless, due to the complex acquisition environment, different degrees of distortion occur frequently in the collected images, affecting the anomaly detection of train. In this paper, a grayscale-information-based segmentation method is proposed for train image registration, aiming at intelligent fault diagnosis of train components. The novelties of the scheme lie in that it is able to correct the distortion promptly, eliminate the distortion of image content accurately, and lay a foundation for the subsequent anomaly detection. At last, the efficacy of the registration scheme is verified for the whole train image (2048×180000 pixels) in the sense of mean square error, normalized cross-correlation matching as well as structural similarity.