Object elimination method using 3D reconstruction with multiple cameras

Abstract

Object erasure techniques are widely used in photo editing and video production to create images that look as if they were never there in the first place by removing unwanted objects and appropriately filling in the background. In recent years, advances in machine learning and deep learning have improved the accuracy of background completion after object elimination. However, the problem is that if the area to be erased is large or the background is complex, the completion results will be unnatural. Furthermore, methods that use only a single viewpoint are limited in the accuracy of completion due to the limited background information that can be obtained. Therefore, we propose a method that utilizes 3D reconstruction for background completion after object elimination. Specifically, we use Instant-NGP for 3D reconstruction and Colmap for estimation of camera parameters to complement the background after erasure.