A deep learning enabled full-automatic visual calibration paradigm for 3D printing system

Abstract To address the printing accuracy challenge in three-dimensional (3D) printing technology, a deep learning enabled full-automatic visual calibration paradigm is developed, which facilitates the widespread use of 3D printers in a low-cost and efficient way. Crucially, the proposed paradigm only requires a photograph of the preprinted test object and a reference object with known dimension using ordinary cameras or even mobile phones for full-automatic calibration. It can enhance the robustness of calibration for micro, thin and light-colored printed objects with the proposed feature enhanced edge detection neural network (FEEDNN). In addition, the shooting angle for calibration is flexible in this paradigm, and then the calibration accuracy can be free from non-standard shooting postures. Therefore, the proposed paradigm has the ability of 3D visual calibration at low cost. A series of printing experiments demonstrate that the printing accuracy is enhanced by 99.02% and calibration efficiency is improved by 11.16% compared to the caliper measurement when printing a 10 mm object using a low-end 3D printing system with millimeter precision, verifying the feasibility and superiority of the proposed paradigm.