A method for constructing depth image super-resolution reconstruction network based on color image guidance

Abstract

The invention discloses a method for constructing a depth image super-resolution reconstruction network based on color image guidance. At present, more and more scene applications require high-quality and high-resolution depth images. Traditional depth image super-resolution methods are not as effective and fast as convolutional neural network-based methods, but most super-resolution convolutional neural network frameworks They all just stack single-size convolution kernels, which cannot extract favorable guiding information from high-resolution color images in the same scene. The multi-scale convolution kernel color image guidance network of the present invention can not only fully explore the high-resolution color image information, extract useful feature images, but also enrich the diversity of depth image features, and fuse these information to perform super-resolution on depth images rate reconstruction. And the idea of ​​recursive learning and residual learning is used to reduce the burden of the convolutional neural network, control the parameters of the network, and improve the reconstruction effect of the depth image.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a method for constructing a depth map super-resolution reconstruction network guided by a color map. Background technique [0002] With the development of technologies for obtaining depth information, such as lidar, time-of-flight (TOF) cameras, 3D structured light, depth images have been widely used in mobile robots, human-computer interaction, human pose estimation and 3D scene reconstruction, etc. However, the depth images obtained from these techniques still cannot meet the practical needs, especially the obtained depth images have low resolution. Therefore, how to reconstruct low-resolution depth images into high-quality and high-resolution depth images has become a research hotspot in the field of computer vision. [0003] In recent years, with the development of deep learning, more and more super-resolution methods based on convolutional neural networks have bee...

AI Technical Summary

Problems solved by technology

[0004] At present, a large number of traditional depth image super-resolution algorithms are based on the color image-guided reconstruction of the same scene. The method of obtaining high-resolution and high-quality color images is relatively mature, but compared with depth images, color images have more texture areas. Color image-guided depth image reconstruction using ordinary convolutional neural networks may have negative effects, so it is necessary to build a network with strong feature extraction capabilities

Images