A device and method for three-dimensional reconstruction of a single frame image based on deep learning

Abstract

The device and method for three-dimensional reconstruction of a single-frame image based on deep learning includes a host computer, a support frame and a high-definition camera on the top, and three 120-degree parallel LED surface light sources are arranged around the high-definition camera contour. The method includes shooting training samples, taking the images illuminated by red, green and blue lights at the same time as the input data required for the training model, taking the gray value of the three images sequentially illuminated by white light, and converting them into single-channel images as the real training model. value, using the deep learning method and the input data to construct a pixel-by-pixel fully connected network model for 3D reconstruction of a single frame image, train the model, use the back propagation algorithm to continuously adjust and optimize the network parameters, and predict the 3D information of the target surface. The result of the network prediction is reconstructed using the photometric stereo algorithm to obtain the three-dimensional information of the surface of the object. The present invention constructs a network model suitable for three-dimensional reconstruction of a single frame image by improving the structure of the network, and the three-dimensional reconstruction of more frame images increases the application scenarios.

Description

technical field [0001] The invention relates to a device for three-dimensional reconstruction of a single-frame image, and the key is a method for three-dimensional reconstruction of a single-frame image based on deep learning, which belongs to the field of three-dimensional reconstruction of a single-frame image Background technique [0002] 3D reconstruction refers to the process of establishing a mathematical model of a 3D object in a real scene in a computer, and it is a popular research direction in the field of computer vision. Compared with two-dimensional images, three-dimensional models can provide depth data of objects, so that they can more comprehensively display the characteristics of objects, so they are widely used in many fields such as computer animation, human-computer interaction, and modern medicine. [0003] However, most of the 3D reconstruction algorithms are 3D reconstruction of multi-frame images. In practical applications, multi-frame images have gr...

AI Technical Summary

Problems solved by technology

[0003] However, most of the 3D reconstruction algorithms are 3D reconstruction of multi-frame images. In practical applications, multi-frame images have great limitations and cannot be reconstructed in real time, and cannot be used for 3D reconstruction of dynamic non-rigid objects.

However, for the traditional single-frame reconstruction method, the accuracy is not high, and the surface details cannot be reconstructed, so it is difficult to apply to scenes requiring high precision

The previous 3D reconstruction algorithm based on deep learning cannot achieve single frame, poor accuracy, poor color robustness, and cannot three-dimensionally reconstruct objects with multiple colors on the surface

Images