Abstract: The problem of current algorithms is lack of timeliness and reliability in bridge crack detection. In this paper, a real-time algorithm on embedded platform is proposed. Firstly, moving average method is used to segment the image coarsely. Then, candidate crack fragments are selected by region growing method using the geometric features of their contours. Finally, a crack merging model with two criteria is built to merge the crack fragments recursively and suppress interference, based on the prior condition of bridge cracks. Experimental results show that the proposed method performs better than several existing methods by 115% at least, especially on extracting hairline cracks and complex cases with uneven illumination and dirty mark. Dealing with an image with 15 megapixel on embedded platform, it costs only 1.73 s.