LLS-LOAM

Lidar Odometry and Mapping with Mutiple Metrics Linear Least Square ICP

Principle

Instead of using non-linear optimization when doing transformation estimation, this algorithm use the linear least square for all of the point-to-point, point-to-line and point-to-plane distance metrics during the ICP registration process based on a good enough initial guess.

How to use

1. Install dependent 3rd libraries:

PCL, Eigen, Glog, Gflags.

If you'd like to use your own data and want to know its absolute projected coordinate, install Proj.

2. Compile

mkdir build
cd build
cmake ..
make 

3. Run

cd ..
# If you'd like to test on kitti dataset
# See below for data preparation
sh script/kitti/kitti_xx.sh
# If you'd like to test on apollo southbay dataset
sh script/apollo/southbay_test.sh
# If you'd like to test on your own dataset, you need to create new shell files and then run it

How to prepare the data

For KITTI dataset, you may refer to Pointcloud Format Transform Tool.

After transforming the point cloud from .bin to .pcd, rename the point cloud folder HDL64.

Horizontally, create a folder called OXTS, and put the ground truth pose .txt file (which can be download from kitti's website) into it.

Then create a pcd file list:

touch file_list.txt
ls HDL64 >> file_list.txt

Then the data is ready for test.

For Baidu Appollo Dataset, you should download the testing data 'Apollo-SouthBay' from Baidu Apollo Data Platform. The data preparation is like kitti's. Remember to use the tool 'rename_number_pcd_files.sh' to regularize the pcd file names (for example, from 1.pcd to 0001.pcd).

For your own dataset, if the point cloud are in pcd format and the pose and imu information are in oxts format, then you can directly use this programme (LoadPcImuGnss Method). Or you need to transform the data youself.

The common folder structure should be

_____base_folder
     |___HDL64
     .   |____0000.pcd
     .   |____0001.pcd
     .   |____.....pcd
     |___OXTS
     .   |____pose.txt
     |___file_list.txt

TO DO LIST

For the loop closure, pose graph optimization and collaboration mapping module, the code has not been released yet.

The workaround is here: CloudControlNet

• Speed up

• Solve the serious Z drift problem

• Use DL to remove the dynamic objects

• Add ROS module

• Tech report or paper

Demo

On KITTI dataset

Yellow: Lidar Odometry Position, Purple: Ground Truth Position

seq 00

seq 01

seq 02

seq 03

seq 04

seq 05

seq 06

seq 07

seq 08

seq 09

seq 10

Quantity Evaluation on KITTI dataset

Apollo SouthBay dataset

On own dataset