Calibration

[munirfarzeen]

Hi,
I want to calibrate two L515 sensors, to make 3D reconstructions from different cameras. How I can do that?

---

Required Info
Camera Model	{ L515 }
Firmware Version	(Open RealSense Viewer --> Click info)
Operating System & Version	Ubuntu 18

[MartyG-RealSense]

Hi @likui01 In a January 2020 Intel conducted a pointcloud stitching demo with nine L515 cameras, Intel demonstrated using a pair of large boards with apriltag images on them to calibrate the positions of the cameras relative to each other.

https://www.youtube.com/watch?v=UzIfn667abE

Intel provided the following additional information.

---

The demo setup consisted of the 9 cameras arranged to give full body coverage, and utilized the hardware sync feature to make sure that the cameras didn’t interfere with each other. In this particular demo, no post processing was utilized to improve the quality of the scan, the only thing done with the raw data was to stitch the 9 point clouds together into one model – so the model you see is the raw data output from the cameras.

In the demo video, you will also see some walls of April tags around the subject. These images were only used in the setup portion of the demo, in order to make sure that the cameras had their positions calibrated relative to each other. No other part of the demo utilized the April tags.

---

It may also be worth exploring whether the RealSense SDK's C++ rs-kinfu application could be used. With this tool, you only need a single camera, and moving the camera around the observed subject progressively builds up the detail on a pointcloud. When you are satisfied you have enough detail on the pointcloud, you can end the data gathering and export the point cloud as a .ply format file. A ply can then be imported into 3D modelling software packages such as MeshLab and Blender for conversion into a solid 3D model.

https://github.com/IntelRealSense/librealsense/tree/master/wrappers/opencv/kinfu

You would likely have to edit the depth stream resolution in the rs-kinfu script to change it from 1280x720 to 1024x768, the maximum supported depth resolution on the L515 model.

https://github.com/IntelRealSense/librealsense/blob/master/wrappers/opencv/kinfu/rs-kinfu.cpp#L190

[munirfarzeen]

@MartyG-RealSense Thank you for the reply.
can you give details on how L515 had their positions calibrated relative to each other using Apriltags? I am using Open3d for 3D reconstruction. But I want to figure out how can I use multiple cameras?

[MartyG-RealSense]

The YouTube video shows that the solid 3D model resulting from the stitching was being viewed in Open3D but does not provide details of the software being used during the scanning process or how the apriltag boards were used for calibration.

There was a similar Open3D project in the link below that uses multiple Azure Kinect cameras to calibrate against apriltags and generate a combined point cloud. You could test whether that project is usable with L515 cameras instead.

https://github.com/stytim/k4a-calibration

If you have not yet purchased the L515 cameras then using RealSense 400 Series cameras instead may work better for your project, as there are more options available for multicam 3D reconstruction with that camera type. For example, there is the project by the CONIX Research Center at Carnegie Mellon that can use up to twenty 400 Series cameras for pointcloud stitching.

https://github.com/conix-center/pointcloud_stitching

The commercial software RecFusion Pro also supports multicam RealSense 3D scanning with 400 Series cameras.

https://www.recfusion.net/products/

[MartyG-RealSense]

Hi @likui01 Do you require further assistance with this case, please? Thanks!

[munirfarzeen]

@MartyG-RealSense, thank you for your response.
I am still unable to calibrate the two L515, it's a work in progress. At this time i don't have an exact question.
I am looking into ways in which i can calibrate using April tags.

[MartyG-RealSense]

Another commercial 3D software tool that supports the L515 camera model is DotProduct Dot3D. It is available in a Lite version and a Pro version for larger scenes that captures up to 160 million points.

https://www.dotproduct3d.com/dot3dlite.html
https://www.dotproduct3d.com/dot3dpro.html

DotProduct has a tutorial for how two separate scans can be registered together using apriltags and Dot3D Pro.

https://desk.zoho.com/portal/dotproduct/en/kb/articles/dot3d-tutorial-registering-multiple-dp-scans-with-apriltags-in-dot3d-pro

[MartyG-RealSense]

Hi @likui01 Do you require further assistance with this case, please? Thanks!

[MartyG-RealSense]

Case closed due to no further comments received.