A 3D-360-degree panoramic image generation method and device

Abstract

The present invention relates to a 3D-360-degree panoramic image generation method and device, a deep learning neural network method is adopted, a left eye view and a right eye view are directly generated through surrounding multi-camera image pictures, and then 3D-eye views are spliced and synthesized. 360-degree panoramic picture, The whole process of obtaining the left eye view and the right eye view is realized by automatic interpolation of appropriate features obtained by automatic training of a network model. no additional optical flow algorithm is needed for calculation; the calculatedquantity is greatly reduced, The method has the advantages that the robustness of panoramic generation is improved, unacceptable artifacts are avoided, the panoramic image generation speed is increased, the image generation process can cope with scenes with complex light rays, close shot targets and insufficient textures, the image quality is high, the efficiency is high, the quality is high, thestability is high, and the requirement of real-time panoramic image collection can be met.

Description

technical field [0001] The invention relates to the technical fields of imaging and image processing, and in particular to a method and device for generating a 3D-360-degree panoramic image. Background technique [0002] With the increasing development of virtual reality (Virtual Reality, VR) technology, the production of VR content has gradually become a short board for the development of the entire industry. 3D-360-degree panoramic video in VR content is an important direction of the video industry, and the performance of its acquisition equipment plays a vital role in the video image quality. [0003] The so-called 3D-360-degree panoramic image means that the field of view of the captured image reaches 360 degrees in the horizontal direction, and at the same time, it contains the left and right eye panoramic views with horizontal parallax, which can be displayed on the left and right eye screens of the VR glasses respectively to achieve a realistic three-dimensional panor...

AI Technical Summary

Problems solved by technology

However, optical flow calculation has always been a difficult problem in the computer vision industry. Optical flow calculation is prone to failure in scenarios such as complex lighting, close-range targets, and poor textures, which will directly lead to errors in the spliced ​​images.

At the same time, the optical flow calculation is to obtain the pixel-level position correspondence of any pixel in the image in another view, and the calculation amount is often very large, which makes it difficult to achieve high frame rate panoramic real-time stitching applications, which also increases the panorama The hardware computing cost of the camera limits the large-scale application of panoramic stitching

Images