Camera Pose Optimization with Genetic Algorithm in Digital Twin

With the increase of interest in digital twin, the way to link dynamic objects into digital twin is also important task as well as static environment. However, there are many situations that the developer can not know all about real environment. This paper focus on finding camera extrinsic parameters(6-DOF camera pose) to fix misalignment between real and digital twin images. Due to the absent of knowledge about function of image $I(\theta)$ given camera parameter $\theta$, this work can’t use gradient descent algorithm known as a powerful optimization method in deep learning. So this paper propose how to apply GA(Genetic Algorithm) for optimizing camera parameters. Fitness function, one of the most important property in GA, is composed of similarity between two images. This paper propose novel approach to calculate similarity using classical feature extraction and matching methods. Even though this work use classic feature extraction method, errors are estimated within 1.42 m for camera position and 1.22° for rotation in all experiments(Averagely 0.744 m and 0.903°). The code of the algorithm can be found in the following link: https://github.com/Kwan-Ho-Kim/Camerapose-optimization.git.