This is an useful toolbox for the developers who need to process the point cloud collected by various types of sensors like LiDAR.

Dependencies

pip install open3d
pip install numpy
pip install opencv-contrib-python

Usage

python tutorial.py

tutorial.py contains several functions, you can comment/uncomment different functions to run the corresponding demos. Here are the functions:

vis_sequential_object() will read the configs and visualize the point cloud with OBB(Oriented Bounding Boxes) in a blocking manner (the non-blocking program is still under developing)
- N: Next Frame
- B: Previous Frame
vis_sequential_sem_inst_motion() will read the configs and visualize the point cloud with Semantic/Instance/Motion labels in a non-blocking manner
- N: Next Frame
- B: Previous Frame
- S: Semantic Mode
- I: Instance Mode
- M: Motion Mode

test_visualizer() contains various kinds of visualizations, you can dive into it.
test_dataloader_visualizer() contains dataloader that can load point cloud from .bin or .pcd
test_kitti_mapper() reads the poses.txt, calib.txt,*.bin,*.label and build a dense map
test_bev() convert the point cloud to BEV(Bird's Eye View) image

The functions of this toolbox include:

dataloader
- load xyz from *.bin
- load xyzi from *.bin
- load xyzi and crop to certain ROI from *.bin
- load semantic/instance label from *.label
- load xyz from *.pcd
- load xyz and downsize from *.pcd
- load xyzi from *.pcd
- load xyzi and downsize from *.pcd
visualize single frame data
- visualize point cloud $[n,3]$
- visualize point cloud with label $[n,3]$ + $[n,1]$
- ……
visualize sequential data
- semantic segmentation
- instance segmentation
- moving object segmentation
- object detection blocking
- object detection non-blocking
mapping with sequential data and poses
- mapping with kitti-like data
visualize kitti point cloud map
lidar-bev transform