kitti_ros

　　A ROS-based player to replay KiTTI dataset.

TODO list

• [x] Publish *.bin Point Cloud in topic /kitti/points_raw _(sensormsgs/PointCloud2).
• [x] Publish ground truth in tracklet_labels.xml.
  • [x] Publish care_objects' 3D OBB (Oriented bounding box) in topic /kitti/bb_raw _(geometrymsgs/PoseArray).
  • [x] Publish as well as /kitti/bb_marker _(visualizationmsgs/MarkerArray) for visualization.
• [x] Publish *.png Image in topic /kitti/img_raw _(sensormsgs/Image).
  • [ ] Publish any Camera 0-3 image.
• [x] Publish *.txt Pose in tf between imu_frame and world_frame.
• [ ] Publish *.txt Calibration in tf between Coordinates.
• [x] KiTTI LiDAR-Camera Fusion, kitti_lidar_camera

How to use

　We name your ros workspace as CATKIN_WS and git clone kitti_ros as a ros package.

# clone source code
$ cd $(CATKIN_WS)/src
$ git clone https://github.com/LidarPerception/kitti_ros
$ cd kitti_ros
# install python's dependencies
$ pip install -r requirements.txt 

# build your ros workspace
$ cd $(CATKIN_WS)
$ catkin build -DCMAKE_BUILD_TYPE=Release

# change Mode for Keyboard Listening Device
$ sudo chmod 777 /dev/input/event3

# [demo] launch kitti_ros's kitti_player with rviz
$ source devel/setup.bash
$ roslaunch kitti_ros demo.launch kitti_data_path:=path-to-your-KiTTI-dataset

　Learn more about parameter: kitti_data_path from Here. We recommend to use quickstart for KiTTI's LiDAR-perception algorithms testing and model training.

# copy quickstart bash scripts
$ cd $(CATKIN_WS)/src/kitti_ros
$ cp quickstart.sh killall.sh ../..

# quick start kitti_ros basic environment and visualization
$ cd $(CATKIN_WS)
$ ./quickstart.sh 

# [option 1] launch kitti_ros's kitti_player for frame-by-frame algorithm testing
$ roslaunch kitti_ros kitti_player.launch kitti_data_path:=path-to-your-KiTTI-dataset
# [option 2] launch kitti_ros's kitti_continue_player for data replay, like model training
$ roslaunch kitti_ros kitti_continue_player.launch dataset_file:=path-to-your-KiTTI-dataset-list-file

# quick exit
$ ./killall.sh

　Learn more about parameter: dataset_file from Here.

　Keyboard Control

• SPACE: Play/Pause KiTTI data replay.
• LEFT ARROW: Play last frame of data.
• RIGHT ARROW: Play next frame of data.

Parameters

• keyboard_file: Keyboard listener is based on Linux input subsystem.
• fps: default 10Hz, the same as LiDAR frequency.
• kitti_data_path: KiTTI raw data directory.
  • default $(find kitti_ros)/../../data/2011_09_26/2011_09_26_drive_0005_sync, that is $(CATKIN_WS)/data/2011_09_26/2011_09_26_drive_0005_sync.

    2011_09_26
    ├── 2011_09_26_drive_0005_sync
    │   ├── image_00
    │   │   ├── data
    │   │   │   ├── 0000000xxx.png
    │   │   │   ├── ...
    │   │   └── timestamps.txt
    │   ├── image_01
    │   │   ├── data
    │   │   │   ├── 0000000xxx.png
    │   │   │   └── ...
    │   │   └── timestamps.txt
    │   ├── image_02
    │   │   ├── data
    │   │   │   ├── 0000000xxx.png
    │   │   │   └── ...
    │   │   └── timestamps.txt
    │   ├── image_03
    │   │   ├── data
    │   │   │   ├── 0000000xxx.png
    │   │   │   └── ...
    │   │   └── timestamps.txt
    │   ├── oxts
    │   │   ├── data
    │   │   │   ├── 0000000xxx.txt
    │   │   │   └── ...
    │   │   ├── dataformat.txt
    │   │   └── timestamps.txt
    │   ├── tracklet_labels.xml
    │   └── velodyne_points
    │       ├── data
    │       │   ├── 0000000xxx.bin
    │       │   └── xxx
    │       ├── timestamps_end.txt
    │       ├── timestamps_start.txt
    │       └── timestamps.txt
    ├── 201?_??_??_drive_0???_sync
    │   ├── ...
    │   └── ...
    ├── calib_cam_to_cam.txt
    ├── calib_imu_to_velo.txt
    └── calib_velo_to_cam.txt

• dataset_file: Only for kitti_continue_player, a list of kitti_data_path line after line.

  • default is $(find kitti_ros)/../../data/training_datasets.txt, as following:

    
    #/home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0001_sync
    /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0005_sync
    /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0014_sync
    /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0017_sync
    #/home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0018_sync
    #/home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0020_sync

  GPF works bad in this dataset

  /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0027_sync

  /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0060_sync

  /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0091_sync

  without ground truth

  /home/gary/Workspace/intern_ws/catkin_ws/data/2011_09_26/2011_09_26_drive_0117_sync

• filter_by_camera_angle: Only care about Camera's angle of view, default true.
• care_objects: default ['Car','Van','Truck','Pedestrian','Sitter','Cyclist','Tram','Misc'], [] means no forground objects.

References

@article{geiger2013vision,
  title={Vision meets robotics: The KITTI dataset},
  author={Geiger, Andreas and Lenz, Philip and Stiller, Christoph and Urtasun, Raquel},
  journal={The International Journal of Robotics Research},
  volume={32},
  number={11},
  pages={1231--1237},
  year={2013},
  publisher={Sage Publications Sage UK: London, England}
}

Thanks

• MarkMuth/QtKittiVisualizer
• yukitsuji/3D_CNN_tensorflow
• strawlab/python-pcl
• utiasSTARS/pykitti