Computer Vision News - December 2019

Open3D for 3D processing 45 print ( "Radius outlier removal" ) cl, ind = voxel_down_pcd . remove_radius_outlier(nb_points = 16 , radius = 0.05 ) As before, the red dots are points that have been removed. It seems that we have a winner and that the radius method is the best for our purpose. Now that we have a clean point cloud, we can save it and move on to some other useful applications of Open3D. ICP Registration There are many applications where we would like to register two point clouds of the same object. To this end, Open3d offers a few registration algorithms: we will use the iterative closest point. ICP algorithm is applied to minimize the difference between two clouds of points, i.e. to register them. It uses an initial rigid body transformation and refines the solution at each iteration by minimizing the distance between the clouds. There are a few variants of this algorithm where Open3d supplies point to point registration and point to plane registration. In the lines that follow, we will use the former one. To apply the algorithm, one can load his two point clouds directly. What we will do is to apply a random rigid transformation on our object with the open3d.geometry. PointCloud.transform method; then, we save it into a separate file. The initial pair of point clouds (in the same coordinates system) looks like this: