Five filter ground methods;
We merge five different ground labels in SemanticKITTI (i.e., road, parking, side walk, other ground and terrain) as ground truth. Table II shows the results of the five different ground filters adopted on the total 11 sequences(00-10) of segmentation data. Considering we only focus on filtering the ground points, the Ground IOU (intersection-over-union of the ground point cloud) is used to evaluate five mentioned filters. The larger the Ground IOU, the higher the accuracy of ground detection. At the same time, filters can also cause false detections, so we also analyzed the detection errors on the three detection targets, e.g., the CarErr, PedErr and CycErr(The error of each class) .The lower the Error, the higher the accuracy of ground detection.
main package : open3d-0.9.0 numpy
ground_removal_ext https://github.com/HViktorTsoi/pointcloud_ground_removal
python filter_ground.py ;
# Change filter ground method by modifying methodnum (in 0-4) in the filterGround function;