Contour-based High-speed Image Registration for Train Fault Diagnosis in Complex Environment

As an important part of image processing, the performance of train image registration imposes a direct impact on the efficiency of image-processing-based fault diagnosis of train systems. Considering the huge size of the train image (e.g., 180, 000 × 2, 048 pixels), it is usually difficult for the traditional image registration methods to meet the high requirements on rapidity and accuracy in complex diagnosis environment. In this paper, a novel contour-based train image registration method is proposed aiming at expediting the diagnosis process and improving the diagnosis accuracy simultaneously. First, the distorted image is preprocessed to eliminate the adverse effects of external interference on the image registration. Second, the contour features of key devices of train are extracted from the preprocessed images. Third, the image is registered by virtue of those features to eliminate the deviation between the distorted and template images. The experimental results show that, compared with the traditional feature point extraction registration method, the registration performance has been improved significantly in the sense that the time consumption is reduced from 517s to 97s, and the registration accuracy is increased from 96.77% to 100%.