Fisheye lenses are widely used in scene surveillance, robot navigation and engineering measurement due to their wide field of view and short focal length. However, the image taken by the fisheye lens has a large degree of distortion, so image correction is the key point of research. In this paper based on the distortion of fisheye image in coal mine underground video surveillance system, the latitude and longitude coordinate method, spherical perspective projection method and panoramic stitching method commonly used in fisheye correction are compared and given an improved algorithm. In the actual survey image correction, preprocessing is an important process, that is, to find the center and radius of the fisheye image. The advantages and disadvantages of the algorithm for finding a circle are also compared.