Will the accuracy of this method be higher than FAST LIO2?

[JACKLiuDay]

Great job for your open-source work. After reading your paper published in TRO, I am curious whether the accuracy of FAST LIVO2 can surpass that of FAST LIO2. If it can, does it outperform FAST LIO2 in all scenarios?

[xuankuzcr]

In our paper, we conducted comprehensive quantitative and qualitative assessments across a wide range of scenarios and the results clearly indicate that FAST-LIVO2 significantly surpasses FAST-LIO2 in all aspects.
In our daily extensive real-world tests, FAST-LIVO2 consistently outperforms FAST-LIO2 by a wide margin.

[JACKLiuDay]

Doctor Zheng, could you provide some useful tools for lidar and camera calibration as a reference? I previously tried some methods in your FAST LIVO work but did not achieve good results. It seems that the biggest issue people will face when trying to reproduce the FAST LIVO2 work is the calibration of the extrinsic parameters.

[JACKLiuDay]

https://gitee.com/gwmunan/ros2/wikis/pages?sort_id=13097190&doc_id=4855084
calibration method

[xuankuzcr]

Thank you for your question! Here are some useful tools for LiDAR and camera calibration:

• Target-based calibration: velo2cam_calibration
• Targetless calibration: direct_visual_lidar_calibration, livox_camera_calib

Additionally, here’s a useful trick for targetless calibration:
You can manually mask the RGB image to enhance the extraction of desired image edges, making it easier to align them with edges from the LiDAR point cloud or intensity image. Below are examples of the original RGB image and the manually masked image.

[JACKLiuDay]

Thank you for your question! Here are some useful tools for LiDAR and camera calibration:

• Target-based calibration: velo2cam_calibration
• Targetless calibration: direct_visual_lidar_calibration, livox_camera_calib

Additionally, here’s a useful trick for targetless calibration: You can manually mask the RGB image to enhance the extraction of desired image edges, making it easier to align them with edges from the LiDAR point cloud or intensity image. Below are examples of the original RGB image and the manually masked image.

Thank you, Dr. Zheng, for your response. We have tried the three methods you mentioned, and among them, dvlc proved to be the most effective, providing a relatively reliable and usable external parameter calibration result. We have also studied the open-source materials from Gundam Company on Bilibili https://www.bilibili.com/video/BV1E3rQYQERf?spm_id_from=333.788.videopod.sections&vd_source=95adb75f5880114d74039abda5341f26 and gained a lot of insights. The tips you mentioned were also very helpful. We have successfully replicated FAST LIVO2 on our own equipment. Additionally, do you have any plans to open-source your other work on improving the computational efficiency of FAST LIVO2? 《 FAST-LIVO2 on Resource-Constrained Platforms: LiDAR-Inertial-Visual Odometry with Efficient Memory and Computation》