A Text-Driven 3D Scene Matching Method Based on Structure Alignment

Abstract

This invention discloses a text-driven 3D scene matching method based on structure alignment, comprising: obtaining instance feature vectors and relation feature vectors of query text and the 3D scene to be matched; counting the number of instances in the query text and the number of instances in the 3D scene to be matched; determining the magnitude of the number of instances in the 3D scene to be matched compared with the number of instances in the query text; matching query instances with scene instances to obtain an instance matching score; extracting relation feature vectors of corresponding instances in the 3D scene and calculating relation matching scores; calculating the consistency score between instances and relations; and performing a weighted summation of the instance matching score, relation matching score, and consistency score to obtain the final matching score between the query text and the 3D scene, thus completing the matching between the 3D scene and the query text. This invention solves the problems of weak structure awareness, low matching accuracy, and insufficient efficiency of traditional methods, and is suitable for large-scale text-driven 3D scene retrieval and matching scenarios.

Images