3D ground segment (GS) instance segmentation method combining two-dimensional segmentation and multi-view voting

Abstract

The application relates to a 3DGS instance segmentation method combining two-dimensional segmentation and multi-view voting, and belongs to the field of three-dimensional scene analysis and processing. The purpose is to solve the boundary blur problem existing in the current three-dimensional Gaussian splash (3DGS) scene instance segmentation. For this purpose, the method first uses a two-dimensional image segmentation model to segment multi-view images of a three-dimensional scene to obtain two-dimensional masks and images of target instances; then uses a motion recovery structure algorithm to obtain sparse point clouds of the three-dimensional scene and camera parameters of each view; based on the sparse point clouds, the camera parameters and the target instance images, uses a 3DGS algorithm for three-dimensional reconstruction to obtain an initial 3DGS model of the target instance; finally, uses a multi-view-based voting strategy to optimize the initial 3DGS model to obtain a 3DGS instance model with clear boundaries and clean surfaces.

Images