A multi-code 3D measurement technique based on deep learning

•To the best of our knowledge, two patterns is current state-of-the-art method for obtaining high-slope absolute phase of complex objects in the FPP system.•The designed ALCNN network has the excellent functions of segmentation and recognition, which can be used to solve the fringe order from only one code pattern.•The proposed MCDL method has high accuracy since the measurement error of a standard ball is 0.0189 mm in FOV of 250 mm × 200 mm.