Research on Low Overlap 3D Point Cloud Reconstruction Algorithm Based on Kinect Camera

3D reconstruction technology covers many fields and is getting closer and closer to public life. In this paper, a complete set of 3D object reconstruction processes based on the Kinect camera is designed by acquiring point cloud data from the Kinect camera and performing 3D reconstruction, in which the acquisition of point cloud data and point cloud registration are the two key problems that need to be solved to realize 3D reconstruction. This paper addresses these two issues. Firstly, the Kinect camera is selected to acquire the point cloud data, and the Kinect camera is calibrated. By analyzing the noise interference of the data acquired by the Kinect camera, it is decided to remove the noise by using the method of bilateral filtering. Secondly, the process of users acquiring point cloud data of target objects with the Kinect camera will be interfered with by various factors, which results in low overlap between 3D point clouds. Therefore, this paper proposes a second-order spatial metric point cloud registration algorithm based on FPFH features, which, compared with the traditional point cloud registration method, is able to have a good registration effect between different overlapping rates, and finally, all the point clouds are registered to the same coordinate system to complete the 3D reconstruction.