Computer Vision News - December 2019

We Tried for You 44 The results of the voxel downsamples are the top figures, while the bottom figures are the results of the uniform. It seems that the downsample is nice, to begin with, but alone it is not enough. To improve it, we apply the next methods on our downsample point clouds. The second method we use is statistical outlier removal. In this method, we opt to remove points that are far from their neighbors compared to the average of the point cloud. We need to choose the parameters of how many neighbors to take and a threshold level based on the standard deviation of the points. We use here an aggressive removal by setting low std value. The code and results are below: print ( "Statistical oulier removal" ) cl, ind = voxel_down_pcd . remove_statistical_outlier(nb_neighbors = 20 ,std_ratio = 0.005 ) Here the red points are the points that have been removed. We can see that this method starts to look better and our main object in the image starts to be clearer. The last method we present is Radius outliers removal. This method removes points that have a few neighbors in a given sphere around them. Here again, we need to tune the number of neighbors and the radius we are looking at.