A 3D Reconstruction Technology of Indoor Scene based on Image Sequence

The 3D reconstruction technology of indoor scenes based on image sequences has always been the focus of research in computer vision. It can be widely used in medical diagnosis, unmanned driving, AR/VR, cultural relics restoration, and other fields. However, due to the complex information and cluttered features of indoor scenes, the existing feature matching algorithms and point cloud registration algorithms still have certain limitations in terms of computational efficiency and matching accuracy. Therefore, this paper firstly adopts a uniform extraction of ORB features method based on octree and a feature matching method based on colour and descriptor distance information and uses the RANSAC algorithm to eliminate mismatched points to obtain matching results with high accuracy. Secondly, this paper adopts a point cloud fine-registration method based on a double threshold constraint. Based on the point cloud normal vector angle threshold constraint, the search of the nearest neighbour point pair in the ICP algorithm is realized through the adaptive distance threshold constraint. Finally, experimental analysis is carried out in a real indoor scene to verify the effectiveness of the proposed algorithm in reconstruction efficiency and accuracy.