The Research On The Point Cloud Reconstruction Based On RGB-D Data

Three-dimensional reconstruction has been an important research direction in the field of machine vision since its inception.It contains various theoretical foundations for optical-mechanical integration and is one of the most popular researches currently.With the rapid development of software and hardware,the related research on 3D reconstruction has gradually begun to apply to production and life.And it is really beginning to be a simultaneous development of production,education,and research.At present,three-dimensional reconstruction has begun to be widely used in many areas such as virtual reality and human-computer interaction,robotics,reverse engineering,and medical training,and has achieved a lot of achievements.This article focuses on the Kinect sensor equipment,and according to the characteristics of the scene depth data and color data acquired by the device.The original point cloud data is further processed and is used to be the research of simplification,registration,and reconstruction on the three-dimensional point cloud.First of all,this article introduces the theory of how to use Kinect on PC and use the interface of OpenNI to acquire 3D point cloud data of real-time scenes.Because there are a lot of noise and invalid points in the acquired raw data,so the OpenCV 2.4.9 library is used for filtering noise and filling holes,which effectively improves the integrity of the point cloud data and provides the possibility for the next research work;This paper proposes a fast registration algorithm based on simplified point cloud data.You first need create a three-dimensional voxel grid for the input point cloud data and calculate the center of gravity of all points in the voxel.Finally,you need use the center of gravity to approximate each voxel in the voxel,so that all points in the voxel can use one the center of gravity indicates that the effect of simplification of the huge point cloud data is achieved;then,the fast ICP algorithm is used to register the simplified point cloud data.This is means that the initial set of points is quickly determined by search algorithm for the simplified point cloud data.The position relationship of the point cloud is initially calculated,and the rough matching process is completed.Based on rough matching,the optimal coordinate transformation is iteratively calculated by the least squares method.And the error is minimized and the pricise matching of the point cloud is finally completed.Finally,the application research method of 3D reconstruction of small-scale indoor scenes is proposed.The first step is to collect a three-dimensional point cloud in a corner of the room.The second step is to detect and matchperform feature point.The last step is to stitch the two three-dimensional point clouds,and complete the interior reconstruction of the three-dimensional scene.The experimental results show that the proposed algorithm can better repair the depth data and achieve point cloud registration at a relatively high speed.And the research work is pushed from the laboratory to the production and living area through indoor reconstruction experiments were completed.Through the research work of this paper,3D reconstruction technology is not only limited to the laboratory stage.After purchasing related inexpensive somatosensory devices,it is possible to experience 3D reconstruction technology in life.It allows more people to participate in the technology.It can provides powerful talents for rapid innovation.