Linear array camera is often used for the moving objects imaging because of the characteristics of high sensitivity, high pixel resolution, and wide dynamic range. In the dynamic detection of running fault image of the multiple units train, the images will stretch or compress in the direction of the train, because the train is not running at an ideal speed through the linear array camera. The number of images taken by the same train at different times is inconsistent, which brings challenges to automatic positioning, identification and automatic fault detection. In order to solve the unaligned problem, we present an block-based image registration method. The image is firstly divided into many sub-blocks, and then feature extraction, matching, and quantization. Each sub-block is corrected in accordance to the pixel distance of feature points. Finally, the correction of the entire images is fulfilled by concatenating the corrected sub-blocks, and the alignment of the target images with the standard image is completed. The experimental result demonstrates that this method has a better performance on alignment for multiple units train sequence images captured by linear array camera. It can accurately positioning the target in the sequence image.