pointcloud_divider

(Updated 2024/06/18)

This is a tool for processing pcd files, and it can perform the following functions:

Dividing point clouds
Merging point clouds
Downsampling point clouds
Generating metadata to efficiently handle the divided point clouds

Supported Data Format

Currently, only pcl::PointXYZI is supported. Any PCD will be loaded as pcl::PointXYZI .

This tool can be used with files that have data fields other than XYZI (e.g., XYZRGB) and files that only contain XYZ.

Data fields other than XYZI are ignored during loading.
When loading XYZ-only data, the intensity field is assigned 0.

Installation

$ git clone https://github.com/MapIV/pointcloud_divider.git
$ cd pointcloud_divider
$ mkdir build
$ cd build
$ cmake ..
$ make

Usage

Select directory, process all files found with find $INPUT_DIR -name "*.pcd".

$ ./scripts/pointcloud_divider.sh <INPUT_DIR> <OUTPUT_DIR> <PREFIX> <CONFIG>

Select individual files

$ ./scripts/divider_core.sh <PCD_0> ... <PCD_N> <OUTPUT_DIR> <PREFIX> <CONFIG>

Name	Description
INPUT_DIR	Directory that contains all PCD files
PCD_0 ... PCD_N	Input PCD file name
OUTPUT_DIR	Output directory name
PREFIX	Prefix of output PCD file name
CONFIG	Config file (default)

INPUT_DIR, PCD_N, OUTPUT_DIR and CONFIG can be specified as both relative paths and absolute paths.

NOTE: The OUTPUT_DIR must already exist. If it does not, an error will occur, and the process will terminate.

Parameter

merge_pcds [boolean]

All PCD files are merge into a single PCD. No divided PCD files are generated.
leaf_size [double]

The leaf_size of voxel grid filter for pointcloud downsampling. The unit is meters [m]. If the value is less than or equal to 0, downsampling is skipped.
grid_size_[xy] [int]

The size of the grid for dividing point clouds. The unit is meters [m]. NOTE: Even if merge_pcds is true, this is used to determine the clusters for downsampling. Therefore, when downsampling without dividing the point cloud, users should not set an excessively large value, such as 100,000. Specifying a large grid size will attempt to load all point clouds into memory and process them at once, which will result in abnormal memory usage.
use_large_grid [boolean]

Pack output PCD files in larger grid directory. When merge_pcds is true, this parameter is ignored. The large grid is fixed at 10 times the size of grid_size_[xy]. For example, if the point cloud is divided into 10m x 10m PCD files, a subdirectory like 00100_00100 will contain up to 100 PCD files.

How the point cloud is processed.

How the PCD file is named

Parameter example

Dividing a single point cloud without downsampling

  use_large_grid: false
  merge_pcds: false
  leaf_size: -1.0 # any negative number
  grid_size_x: 20
  grid_size_y: 20

Downsampling and merging divided point clouds into a single file

  use_large_grid: false
  merge_pcds: true
  leaf_size: 0.2
  grid_size_x: 20
  grid_size_y: 20

Metadata YAML Format

The metadata file should be named metadata.yaml. It contains the following fields:

x_resolution: The resolution along the X-axis.
y_resolution: The resolution along the Y-axis.

Additionally, the file contains entries for individual point cloud files (.pcd files) and their corresponding grid coordinates. The key is the file name, and the value is a list containing the X and Y coordinates of the lower-left corner of the grid cell associated with that file. The grid cell's boundaries can be calculated using the x_resolution and y_resolution values.

For example:

x_resolution: 100.0
y_resolution: 150.0
A.pcd: [1200, 2500] # -> 1200 <= x <= 1300, 2500 <= y <= 2650
B.pcd: [1300, 2500] # -> 1300 <= x <= 1400, 2500 <= y <= 2650
C.pcd: [1200, 2650] # -> 1200 <= x <= 1300, 2650 <= y <= 2800
D.pcd: [1400, 2650] # -> 1400 <= x <= 1500, 2650 <= y <= 2800
...

LICENSE

This repository is under BSD-3-Clauses license