The work aims to design a cylindrical label unwrapping algorithm based on cylinder fitting to solve the problem that the monocular camera can not obtain all the valid information of the label during detection due to limited field of view. Firstly, the ideal position of the cylindrical label was fitted from the external parameters of the camera obtained from the camera calibration, and the height of the cylindrical label unwrapping image was determined. The fuzzy membership function was determined according to the extreme position of the cylinder in the field of view of the camera, and the edge points of the extracted cylindrical label were filtered. Finally, the position of the cylindrical product was determined according to the world coordinates of the edge points. Pixel values were extracted from monocular camera image and assigned to corresponding points in the cylindrical label unwrapping image, and finally, the label unwrapping image was obtained. If there were high requirements on the quality of stitching, the NCC matching algorithm could be used to adjust the stitched image to achieve better results. This method could achieve the rapid stitching of multiple cylindrical label images and the label unwrapping image was obtained. The cylindrical label image of a radius of about 30 mm could be stitched at a speed of 110 ms, which laid the foundation for online detection of label quality. This method can quickly and accurately unwrap the side label of cylindrical product.