Skip to content

LiDAR_preprocessing.md

Latest commit

 

History

History
140 lines (111 loc) · 4.36 KB

LiDAR_preprocessing.md

File metadata and controls

140 lines (111 loc) · 4.36 KB
Raw

LiDAR preprocessing

  • Seperate ground and object in LiDAR data

  • Core Flow

              ROS 2 
                │
                │ /kitti/point_cloud
                │ 
                ▼
 ┌─────────────────────────────────────────┐
 │ sensor_msgs/msg/PointCloud2             │
 │              │                          │
 │              ▼ (pcl_conversions)        │
 │       PCL PointCloud                    │
 │              │                          │
 │              ▼ (pcl::VoxelGrid)         │
 │       VoxelGrid Downsample              │
 │              │                          │
 │              ▼ (pcl::SACSegmentation)   │
 │   RANSAC Plane Segmentation             │
 │              │                          │
 │      ┌───────┴───────┐                  │
 │      ▼               ▼                  │
 │  [Success]       [Failure] ───┐         │
 │      │                        │         │
 │      ▼ (pcl::ExtractIndices)  │         │
 │  ┌───────────────┐            │         │
 │  │ Ground Points │            │         │
 │  └───────┬───────┘            │         │
 │          │                    │         │
 │          ▼                    │         │
 │  ┌───────────────┐            │         │
 │  │ Obstacles     │            │         │
 │  └───────┬───────┘            │         │
 └──────────┼────────────────────┼─────────┘
            │                    │
            ├────────────────────┘ (Drop / Log Error)
            ▼
 ┌─────────────────────────────────────────┐
 │          ROS 2 Publishers               │
 │                                         │
 │  /lidar_preprocessing/ground_pcd        │
 │  /lidar_preprocessing/object_pcd        │
 └─────────────────────────────────────────┘
  • Receive input from LiDAR, subscrib topic /kitti/point_cloud
  • Covert ROS2 PCL to PCL for sepearte ground and object
  • Downsample voxel
  • Use RANSAC plane segmentation method to seperate ground and object
  • Publish ground PCL and object PCL
  1. Create a node for LiDAR preprocessing
cd lidar_camera_perception/src
code lidar_preprocessing.cpp
  1. Modify CMakeLists.txt to compile LiDAR_preprocessing
  • Add required package
find_package(PCL REQUIRED COMPONENTS common filters io segmentation sample_consensus) # PLC
find_package(pcl_conversions REQUIRED)
  • CPP executable
add_executable(lidar_preprocessing src/lidar_preprocessing.cpp)
  • PLC directory, definitions, library for PCL
# Include PCL directories
target_include_directories(lidar_preprocessing PUBLIC
  ${PCL_INCLUDE_DIRS}
)

# Compile definitions for PCL
target_compile_definitions(lidar_preprocessing PUBLIC
  ${PCL_DEFINITIONS}
)

# Link ROS2 and PCL libraries
target_link_libraries(lidar_preprocessing ${PCL_LIBRARIES})
  • ament macro to link standard ROS2 dependencies
ament_target_dependencies(lidar_preprocessing
  rclcpp                        # for ROS2 C++ client library
  sensor_msgs                   # for PointCloud2 messages
  pcl_conversions               # for converting between PCL and ROS2 PointCloud2 messages
)
  • Installion configuration
install(TARGETS lidar_preprocessing
  DESTINATION lib/${PROJECT_NAME}
)
  1. Build
colcon build --symlink-install --packages-select lidar_camera_perception
  1. Test
  • Run object detection
ros2 run lidar_camera_perception lidar_preprocessing
  • Run input
ros2 bag play kitti_dataset_0009
  • rviz2 for visualization
rviz2

lidar_preprocessing