An efficient parameter update method of 360-degree VR image model

摘要

Recently, with the rapid growth of manufacture and ease of user convenience, technologies utilizing virtual reality images have been increasing. It is very important to estimate the projected direction and position of the image to show the image quality similar to the real world, and the estimation of the direction and the position is solved using the relation that transforms the spheres into the expanded equirectangular. The transformation relationship can be divided into a camera intrinsic parameter and a camera extrinsic parameter, and all the images have respective camera parameters. Also, if several images use the same camera, the camera intrinsic parameters of the images will have the same values. However, it is not the best way to set the camera intrinsic parameter to the same value for all images when matching images. To solve these problems and show images that does not have a sense of heterogeneity, it is needed to create the cost function by modeling the conversion relation and calculate the camera parameter that the residual value becomes the minimum. In this article, we compare and analyze efficient camera parameter update methods. For comparative analysis, we use Levenberg–Marquardt, a parameter optimization algorithm using corresponding points, and propose an efficient camera parameter update method based on the analysis results.