I am working on a setup with two ZED 2i cameras mounted externally at fixed positions around my workspace. I would like to determine the relative position and orientation (extrinsic parameters) of the two cameras accurately so that I can use their data in a common coordinate system and potentially fuse the 3D/depth information.
I am currently exploring ZED360 calibration and would like to understand the recommended workflow for this type of setup.
Has anyone here successfully performed external/extrinsic calibration between two ZED 2i cameras?
I would particularly appreciate guidance on:
What calibration method/tool did you use: ZED360, OpenCV, or another approach?
What was your physical setup and camera arrangement?
How did you ensure good calibration accuracy and repeatability?
Do the cameras need to remain completely fixed after calibration?
How do you validate that the resulting relative positions and orientations are accurate?
Is ZED360 the recommended approach if I want to use both cameras in a common 3D coordinate system?
Are there any examples, sample code, or best practices for a two-camera ZED 2i setup?
Connect both cameras to the same machine, open ZED360, run Auto Discover, then have one person walk slowly through the whole covered area. The tool refines the camera poses from body keypoints roughly every ten seconds, so accuracy comes from covering the full space and from the two views overlapping.
The cameras must stay fixed. The configuration is valid only as long as the cameras don't move, so any repositioning means recalibrating.
To validate the result and to view both point clouds in one frame, the sensors_api/managed sample in the ZED SDK (5.2 and above) reads the same poses and draws every camera's cloud in a single scene.
Note that Fusion merges object detection and body tracking across cameras, it does not merge depth maps into one depth image.
Thank you, Tristan. This is very helpful and clears up my doubts. I’ll proceed with ZED360, ensure sufficient overlap between the two cameras, and keep the cameras fixed after calibration. I’ll also check the sensors_api/managed sample for validating the poses and viewing both point clouds. Thanks again for the detailed guidance!