pointcloud-toolkit

The toolkit provides scripts for exploratory data analysis, converting ROS bag files, and visualizing point cloud data.

Data Structures

The scripts require that data be organized as follows within a directory (e.g., custom).

├── custom
│   ├── points
│   │   └── 001.npy
│   ├── labels
│   │   └── 001.txt

The points folder should contain the point cloud data files in .npy format. Each line within these files represents a point with the following format: x y z intensity. Here:

• x, y, z are the spatial coordinates,
• intensity represents the reflectivity strength of the surface.

The labels folder should contain the label data files in .txt format. Each line within these files defines a bounding box with the following format: x y z dx dy dz yaw track_id type. Here:

• x, y, z are the centroid coordinates of the bounding box,
• dx, dy, dz denote the dimensions of the box (length, width, height),
• yaw indicates the rotation around the vertical axis,
• track_id is a unique ID used for tracking objects across frames,
• type describes the class of the object.

Each point cloud and its corresponding label from each frame must share the same base filename, differing only in their file extension. This ensures that each point cloud data file is correctly associated with its corresponding label data.

ROS Conversion

Use rosbag2npy.py to convert ROS bag files into NumPy arrays.

python rosbag2npy.py --bag_path ${rosbag_path} --output_path ${output_path} --lidar_topic ${lidar_topic}

• --bag_path: Path to the ROS bag file.
• --output_path: Directory where the output .npy files should be saved.
• --lidar_topic: ROS topic name that contains the lidar data.

Data Analysis

Use eda.py to perform exploratory data analysis on point cloud data.

python eda.py --seq_list ${sequence_name} --seq_path ${sequence_path}

• --seq_list: A list of folder names, where each name corresponds to a sequence that is analyzed.
• --seq_path: Directory path that contains the folders for the sequences specified in the --seq_list.
• --eda_name: Name of the EDA output (default: custom).
• --sample: Number of bounding boxes to sample for analysis (default: 500). If this parameter is not specified, all bounding boxes in each sequence will be analyzed.
• --bins: Method to calculate histogram bins (default: sqrt).
• --xmax: Maximum value of the x-dimension (default: 2000).

Visualization

Use vis_seq.py to visualize point cloud data sequences.

python vis_seq.py --points_path ${points_path} --remove_intensity

• --points_path: Path to the directory containing the point cloud data.
• --remove_intensity: Draw point cloud data without intensity values.

Use vis_gt.py to visualize ground truth bounding boxes in point cloud data.

python vis_gt.py --data_path ${data_path} --remove_intensity

• --data_path: Directory path containing both points and labels folders. If this parameter is used, --points_path and --labels_path will be set to the points and labels subfolders within this directory, respectively.
• --points_path: Specific path to the point cloud data file or directory. This is only needed if --data_path is not used.
• --labels_path: Specific path to the labels file or directory. This is only needed if --data_path is not used.
• --ext: File extension for point cloud data files (default: .npy).
• --remove_intensity: Draw point cloud data without intensity values.
• --use_tracking: Draw traces by using tracking IDs from labels.
• --bins: Filter bounding boxes based on the number of points inside. If one number is provided, it sets the range between 0 and this number. If two numbers are provided, it sets the range between these two numbers.
• --all: Used in conjunction with --bins. Draw selected bounding boxes in red color with other boxes in green.