基于RGB-D传感器的室内三维重建方案

In this paper, a fast real-time dense reconstruction algorithm based on image key frames and RGB-D sensors is proposed. The traditional SFM-based reconstruction algorithm has problems such as insufficient parallax and excessive calculation, and it is difficult to perform real-time reconstruction in large terrain and large scenes. The problem of dense reconstruction is based on the visual-inertial odometry (VIO) to obtain the pose information and the depth information obtained by the RGB-D sensor. First, the point cloud is quickly spliced and preliminarily filtered; For the problem of excessive noise, large volume and misIndoor 3D Reconstruction method based on RGB-D sensorsing texture in the point cloud map, the dense point cloud map is further processed, and a classical Poisson reconstruction algorithm is used to construct a scene grid map with a clear physical structure and light weight; for dense point cloud Due to the lack of map texture and the large cumulative error of VIO pose, the pose is globally optimized using the assumption of luminosity consistency, and then the rich textures of the RGB image are mapped onto the grid map through the pose, and finally a lightweight , 3D reconstruction results of scenes with rich textures.