Skip to content

Latest commit

 

History

History
148 lines (117 loc) · 6.75 KB

README.md

File metadata and controls

148 lines (117 loc) · 6.75 KB
Raw

SC-F-LOAM

ROB-530-Final-Project

Prerequisites (dependencies)

  • We mainly depend on ROS, Ceres (for A-LOAM), and GTSAM (for pose-graph optimization).
    • For the details to install the prerequisites, please follow the A-LOAM, LIO-SAM, and F-LOAM repositiory.
  • The below examples are done under ROS melodic (ubuntu 18) and GTSAM version 4.x.

How to use?

  • First, install the abovementioned dependencies, and follow below lines.
    mkdir -p ~/catkin_scfloam_ws/src
    cd ~/catkin_scfloam_ws/src
    git clone https://github.com/zirui-xu/ROB-530-Final-Project.git
    cd ../
    catkin_make
    source ~/catkin_scfloam_ws/devel/setup.bash
    roslaunch scfloam scfloam_velodyne_HDL_64.launch # Import data after launching it
  • Output will be saved to save_directory in scfloam_velodyne_HDL_64.launch. Please modify it to the directory you want. Save utility follows SC-LIO-SAM's save utility.

Some problems you might encounter

  • catkin_make fail: make it until success (in my case, 3 times) if you install everything well.

How to import data?

Raw data way:

Download data from Kitti website to YOUR_KITTI_DATASET_DIRECTORY.

- Download odometry data set (grayscale, 22 GB)
- Download odometry data set (velodyne laser data, 80 GB)
- Download odometry ground truth poses (4 MB)

The directory should form like

| YOUR_KITTI_DATASET_DIRECTORY 
    |-- results
        |-- 00.txt
        |-- 01.txt
        |-- ...
    |-- sequences 
        |-- 00
            |-- calib.txt
            |-- times.txt
            |-- image_0
                |-- 000000.png
                |-- 000001.png
                |-- ...
            |-- image_1
                |-- 000000.png
                |-- 000001.png
                |-- ...
            |-- velodyne
                |-- 000000.bin
                |-- 000001.bin
                |-- ...
        |-- 01
        |-- ...

Modify kitti_helper.launch to fit your data (e.g., dataset_folder).

    source ~/catkin_scfloam_ws/devel/setup.bash
    roslaunch scfloam kitti_helper.launch

Note: You can also generate rosbag by running the above command. Just change parameters of kitti_helper.launch (i.e., set to_bag to true and output_bag_file to YOUR_ROSBAG_DIRECTORY).

Rosbag way:

Play the wrapped kitti data in XXX.bag. Here you can download some preprocessed rosbags.

    rosbag play YOUR_ROSBAG_DIRECTORY/XXX.bag

Note: you should run roscore in the other terminal before running the above command.

Results:

Video

image

The result trajectory generated for KITTI 00, KITTI 02, and KITTI 05 are shown as below, see "How to evaluate?" for evaluation. The dashed line is the ground truth of the map, the blue line is the estimate of F-LOAM, and the red line for SC-F-LOAM.

The following histogram is comparison of A-LOAM, SC-A-LOAM, F-LOAM, and SC-F-LOAM on KITTI 00, KITTI 02, and KITTI 05. ATE and ARE are in absolute value, and ARE is a unitless estimate.

image

The following chart shows the ATE and ARE of the four methods in KITTI 00, KITTI 02, and KITTI 05 specifically.

Method Metric KITTI 00 KITTI 02 KITTI 05
SC-A-LOAM ATE 1.158887 103.368896 0.712823
SC-A-LOAM ARE 0.028809 0.476638 0.016656
SC-F-LOAM ATE 1.465621 3.483975 1.32898
SC-F-LOAM ARE 0.02746 0.047693 0.018149

How to evaluate?

  • First, you need to put the output file generated by SC-F-LOAM to certain folders. odom_poses.txt, optimized_poses.txt, times.txt, convert.m (in project folder), and the ground truth (00.txt for Kitti00) should be arranged in the directory like:
| YOUR_PROJECT_OUTPUT_DIRECTORY 
    |-- Kitti00
        |-- SC-F-LOAM
            |-- odom_poses.txt
            |-- optimized_poses.txt
            |-- times.txt
        |-- 00.txt
        |-- convert.m
    |-- Kitti01
    |-- ...

  • After generating corresponding files, you need to open convert.m file and change the name of file for 'reading time and sample ground truth accordingly' to 'SC-A-LOAM/times.txt' or 'SC-A-LOAM/times.txt' if you want to estimate SC-A-LOAM or SC-F-LOAM.

  • Then run convert.m file to generate a ground_truth.txt file from the '00/02/05.txt' file of ground truth. This is because the frames and the x, y and z dimensions of original ground truth are not the same as output estimates.

    • The function of convert.m is to change the dimension and number of frames of ground truth file to match the output file of F-LOAM and SC-F-LOAM.
    • It changes x, y, and z dimension of ground truth by z to x, -x to y, and -y to z, so that dimensions and directions match.
    • Then it shuffles the ground truth frames according to times.txt which represents actual time of the vehicle for each output frame.

  • Finally, you can install “evo”, an evaluation tool, from https://github.com/MichaelGrupp/evo or by pip install evo or other methods you prefer. Then you can use evo for evaluation according to its instructions. Some sample commands for evaluation are shown as below:

    • If you want to compare trajectories against ground truth, you can use the following command:
    evo_traj kitti SC-F-LOAM\odom_poses.txt SC-F-LOAM\optimized_poses.txt --ref=ground_truth.txt -p --plot_mode=yx
    • If you want to estimate error, use
    evo_ape kitti ground_truth.txt SC-F-LOAM\odom_poses.txt -va --plot --plot_mode yx --save_results SC-F-LOAM/odom_poses.zip
    • To make detailed comparison use
    evo_res SC-F-LOAM\odom_poses.zip SC-F-LOAM\optimized_poses.zip -p --save_table SC-F-LOAM/comparison.csv

License

Our code is made publicly available for the benefit of the community