A 3D Reconstruction System for Large Scene Based on RGB-D Image

As an important research topic in the field of computer vision, 3D modeling of complex scene has an extensive application prospect. RGB-D sensor has been widely used to obtain the depth information in recent years. However, the existing processing system is simply suitable for small-scale scene modeling. In order to develop a better algorithm for large-scale complex scene modeling, this paper builds a 3D scene reconstruction system based on RGB-D images, achieving a better performance in accuracy and real time. SIFT algorithm is first to extract key points to match the descriptors between consecutive frames. By converting into three-dimensional space through the intrinsic matrix, the effective pixel points in the images are then reintegrated to establish the spatial point clouds model which is finally optimized by RANSAC algorithm. The experiments are based on the public database and propose the solution of the problems in the system, which provides a platform for basic research work.