Abstract: To enhance the flattening quality and efficiency of plates and realize intelligent flattening, in this work, machine vision is utilized in place of manual recognition to realize accurate recognition of the three-dimensional (3D) contour of warped plates. Modeling research on the relevant processes of point cloud recognition, surface fitting, and flattening process is conducted. A structured light camera is utilized to obtain the point cloud data of the 3D contour of warped plates, and the recognition accuracy is found to meet the requirements of the flattening of the warped plate. Preprocessing of the point cloud data is conducted. The ground point cloud is segmented from the point cloud of the warped plate based on a shape model method. The inner outline of the ground point cloud is extracted, and afterward, the warped plate at any position is converted into a new pose, with the plate center serving as the origin of the camera coordinate system and the plate edge being parallel to the coordinate axis. It provides a foundation for acquiring the warped plate distribution in the coordinate system of the flattening machine, but it also facilitates subsequent point cloud denoising, which is conducted using conditional filters and distance-based techniques. Finally, the point cloud is simplified to enhance the computational efficiency using the principle of sparse sampling points. The 3D surface of the warped plate is reconstructed using the least-squares method. The curvature of the warped plate is computed using the surface theory in differential geometry, and the positions of the warped plate’s pad and pressing points are optimized based on the 3D surface features. Based on the three-point bending leveling theory, the screw-down force and displacement of warped plate are obtained. In comparison to the conventional modeling research of the flattening process based on the two-dimensional warping contour, we establish a reconstruction method of the warped plate’ 3D contour and a curvature model and determine the optimization principles for the positions of the pad plate and pressing point thereby providing theoretical models for intelligent flattening of plates. A finite element model of the flattening of the warped plate demonstrates that the deviation between the simulation and theoretical results for the flattening force is less than 5.4%, and the residual warping height is about 1 mm, indicating the reliability of the theoretical model. The deviation between the simulation and theoretical screw-down force is about 2.21%, and the unevenness decreased from 17.2 to 3.28 mm·m^?1, demonstrating that the models and methods are relatively accurate and feasible.