The safety is very important for urban rail transit. So it is the necessary to detect the crack of subway tunnel to keep the secure of the line. In traditional way, relevant staffs are asked to go into the tunnel to detect the crack by naked eye. But this method is too inefficiency to meet the actual demand. In this paper, we propose an algorithm to detect the crack based on the image processing. Compared with other algorithm, this algorithm simplifies the image preprocessing procedure, and use the block binaryzation to replace the traditional method of binaryzation. Then extraneous noisy pixels are removed from the image according to the differences of characteristics between noise and crack. Finally, crack can be detected automatically.