Research On 3D Modeling Rapid Reconstruction Of Real Scene

Three-dimensional reconstruction technology is the computer graphics,ancient cultural relics protection,VR/AR and other cutting-edge areas of key technologies,has been widely used in various fields.Aiming at the high cost of 3D reconstruction,the long modeling period and cumbersome operation,a threedimensional reconstruction based on Kinect is proposed,which can reconstruct the surrounding objects or scenes quickly,easily and cheaply.The main contents are as follows:(1)In the process of obtaining the cloud data,the data of the point cloud data obtained by Kinect sensor is large and contains noise points,PassThrough,Statistical OutlierRemoval and VoxelGrid under PCL are used to remove the background data and remove the outliers and descramble processing respectively.After filtering the cloud data volume and noise significantly reduced,but the details of the point cloud has no effect,you can facilitate the follow-up registration operation.(2)A new method based on improved KinectFusion is proposed.The main improvement is the following two points:1)In the process of registration of point cloud,KinectFusion reconstruction method is adopted ICP registration algorithm,the algorithm requires a higher initial position,and the registration accuracy is low.Aiming at this problem,a point cloud registration algorithm based on SAC-IA and improved ICP is proposed.The algorithm first calculates the FPFH feature of the point cloud surface,and then uses the SAC-IA algorithm to compute the transformation matrix of the corresponding point pairs.Point cloud for matrix transformation to complete initial registration.Finally,the improved ICP algorithm is used for accurate registration.The improvement is mainly composed of two parts: the kd tree(k-dimension tree)is used to improve the search speed of the corresponding point pairs in the algorithm.The Euclidean distance threshold and the point cloud direction vector threshold are used to reject the corresponding pairs of errors.The algorithm effectively improves the registration accuracy of point cloud.2)KinectFusion method to rebuild the scene model and no texture information,can not be real on the scene to reproduce.The system uses the texture mappingmethod based on color fusion to paste the rich texture information on the scene model surface,and obtain a more intuitive and realistic three-dimensional scene model.The three-dimensional scene model reconstruction method proposed in the system can make the ordinary users use the cheap Kinect sensor to connect the computer to the simple scene three-dimensional reconstruction,compared with the traditional reconstruction method to reduce the cost of three-dimensional modeling and complexity.The three-dimensional scene model of the final reconstruction has high precision and good texture information,which has certain application prospect in actual production and life.