Performance Evaluation of Different Optimization Techniques for Image-based Camera Localization

Abstract

In this study, the performance evaluation of different optimization methods in calculating the change of camera movement, in other words the camera position and angle, was made on images taken from different views with the same camera. For this purpose, gradient-based and population-based optimization techniques have been investigated by using the differentiable rendering methods that have been proposed in recent years for the 3D point cloud created over depth information. It was found that while gradient-based methods are more successful on close movements, population-based methods were more successful in distant changes.