Abstract: Currently the distribution of cameras at fully mechanized working face is dispersed,and the mine intelligent applications need more big scene and high-resolution images. The wide-angle vision camera and panoramic camera for underground mines try to restore the broad field of vision for fully mechanized working face as much as possible. How- ever,due to the equipment from a single point in image acquisition,image distortion will appear,and even result in los- ing image fidelity. Therefore,it is necessary to do more visual technology research on the global coordinate system. This paper introduces the installation and design of the image acquisition device to make it meet the requirement that the collected images are overlapped to a certain extent and the features are stable. An improved histogram equalization method is proposed by compressing or stretching the gray scale range of the working face image,which can improve the image detail reflection under low illumination and the suppression effect under partial strong light,so as to improve the identification of underground environment and optimize the image quality. Surf algorithm is used to match and extract feature points,allocate the main direction of feature points and complete the description of feature points. Aiming at the situation that there are many mismatched feature points in this algorithm,RANSIC algorithm to eliminate mismatched feature points is studied,and the number of valuable feature points is increased by means of corner point detection and pairing. Finally,the perspective matrix expression and the global coordinate system are established through the coordi- nate point coordinates to complete the image splicing,and the weighted smoothing method is adopted to remove the cracks in the spliced images,so as to realize the fast and effective fusion of multi-vision images. Experiments show that this method can effectively enhance underground mine image details,remove underground mine image mismatching feature points,quickly eliminate stitching cracks,and quickly and accurately complete two or three images stitching fusion.