LiDAR_Camera_Calibration_Preprocess

This repository contains MATLAB and Python tools to extract and synchronize point clouds and images from a rosbag for extrinsic calibration.

1. Prerequisites

1. ROS (tested on Kinetic and Melodic)
2. Python 2.X (tested on 2.7.17)
3. MATLAB (tested on 2020a and 2021b)

2. Data Extraction

2.1 Point Cloud

Use pcl_ros to extract point clouds from a rosbag:

rosrun pcl_ros bag_to_pcd <input_file.bag> <topic> <output_directory>

2.2 Image

1. Change the following lines in config.ini:

   ; Parameter Setup
   [DEFAULT]
   ; ; Change rosbag path here.
   ; bag_path = <your roabag path>
   ; ; Change image topic here.
   ; image_topic = <your image topic>
   ; ; Change image destination folder path here.
   ; image_path = <your image destination folder path>
   ; ; Change cropped image destination folder path here.
   ; cropped_path = <your cropped image destination folder path>
   ; ; Change cropped resolution from bottom here.
   ; cropped_resolution = <your cropped resolution from bottom>

2. Use extract_image.py to extract images from a rosbag (Python 2.X):

   python extract_image.py

3. Use crop_image_bottom.py to crop images from bottom (Python 2.X):

   python crop_image_bottom.py

3. Data Synchronization

1. Change the following lines in main.m:

   %% Determine Folder Path
   path = "<your root folder path>"; % Change folder path here.
   path_img = strcat(path, "img/*.jpg"); % Place images       in ./img/ folder
   path_pcd = strcat(path, "pcd/*.pcd"); % Place point clouds in ./pcd/ folder

2. Run main.m to synchronize timestamps.

3. Check synchronized data in ./img_sync/ and ./pcd_sync/ respectively.

4. Extrinsic Calibration

After extracting and synchronizing data from the rosbag, use CameraCalibrator app in MATLAB to calculate the intrinsic of the camera. Later, use the method mentioned in the following paper¹ to calibrate the extrinsic between LiDAR and camera.

5. TODO

• Read config file to get parameters.
• Check and create destination folder.
• Add progressbar.
• Use Python to extract both point cloud and image.
• Use Python to synchronize point cloud and image messages.
• Manually select images and corresponding point cloud.
• Release and maintain code for extrinsic calibration (All credit goes to the authors of the paper above).

6. Reference

1. pcl_ros
2. ROS Answers
3. CSDN Blog
4. CSDN Blog
5. CSDN Blog
6. CSDN Blog
7. Stack Overflow
8. Stack Overflow

Footnotes

1. Xie S, Yang D, Jiang K, et al. Pixels and 3-D Points Alignment Method for the Fusion of Camera and LiDAR Data[J]. IEEE Transactions on Instrumentation and Measurement, 2019, 68(10): 3661-3676. [LINK] ↩