Three-Dimensional Reconstruction Method with Parameter Optimization for Point Cloud Based on Kinect v2

Kai-zhang WANG, Tong-kai LU, Qi-hang YANG, Xi-hao FU, Ze-hong LU, Bo-lun WANG, Xin JIANG

Abstract


Three-Dimensional (3D) reconstruction is a significant part in the field of computer vision. In this paper, we use Kinect v2 to obtain the 3D point cloud. As a consumer used 3D sensor, Kinect has numerous advantages such as low price, relative high dot per inch (DPI) and frames per second (FPS) and strong robustness. During the experiment, the object was placed on a turntable spinning and stop for every 12°, collecting 30 images of point cloud in total. Then we used Iterated Closest Points (ICP) algorithm to calculate the optimal rotation matrix and translation matrix to match all point cloud into the same coordinate. After that, three algorithms including Statistical Outlier Removal, Movement Least Squares (MLS) and Voxelgrid were used to reduce the noise. At last, we applied greedy projection algorithm to generate the triangulate mesh. During the procession, we observed the relationship between parameters and outcome and drew pictures to have the data more visible, which gave out the optimal parameters in return.

Keywords


3D Reconstruction, Parameter optimization, Kinect v2


DOI
10.12783/dtcse/cscbd2019/30058

Full Text:

PDF

Refbacks

  • There are currently no refbacks.