Accurate and fast large-depth-of-field three-dimensional reconstruction method based on deep learning

The invention relates to an accurate and fast large-depth-of-field three-dimensional reconstruction method based on deep learning, and belongs to the technical field of computer intelligent vision. Comprising the following steps: constructing a depth-of-field expansion convolutional neural network based on deep learning design, collecting an original fringe picture of a to-be-measured object through a three-dimensional measurement system, obtaining a high-precision wrapped phase, and reconstructing three-dimensional information according to the high-precision wrapped phase. According to the method provided by the invention, a high-precision wrapped phase can be obtained in a larger measurement depth of field by utilizing three fringe images with different phase shifts shot by equipment at a fixed focal length. In the process, measurement errors caused by the projector and the camera can be obviously reduced, and high-performance three-dimensional reconstruction can be realized in a large depth-of-field scene