Hi ! I would like to compute a depthmap and a pointcloud with a custom setup using a FFC-4P and two 0V9282 mono cameras. The eeprom is flashed with the stereo calibration (intrinsics + extrinsics) I am able to get a disparity map by creating a stereoDepth node, but I would like to display a point cloud to get a better idea of what's going on. I could not get [this](https://github.com/luxonis/depthai-experiments/tree/master/gen2-pointcloud) to work (either with the FFC or with a OAK-D pro). [depthai-viewer](https://github.com/luxonis/depthai-viewer) worked perfect with the OAK-D pro, but returned an error when plugging in the FFC: > stream_enabled=True name='Right' is not configured. Couldn't create stereo pair. What would be the correct procedure for producing a point cloud from a custom setup ? Additionnally, we want to try to add the IR projector module, I saw that our ffc needs to be R5 or newer, how can I check that ? [upl-image-preview url=https://discuss.luxonis.com/assets/files/2023-09-13/1694617214-187969-image.png] Thanks ! Antoine

@"apirrone"#p12043 Could you give me what you consider to be acceptable calibration results ? I got * right Reprojection Error: 0.516754 * left Reprojection Error: 0.473438 * Average Epipolar Error is : 0.16123293063130242 Which was good for visual stereo (in VR) but maybe it is not good enough for computing depth maps ? Thanks !

Hi Erik, The rectified streams seem ok [upl-image-preview url=https://discuss.luxonis.com/assets/files/2023-09-18/1695028108-675706-capture-decran-du-2023-09-18-11-07-37.png] [upl-image-preview url=https://discuss.luxonis.com/assets/files/2023-09-18/1695028231-809344-capture-decran-du-2023-09-18-11-10-25.png]

FFC 4P depthmap / point cloud

apirrone

Hi !

I would like to compute a depthmap and a pointcloud with a custom setup using a FFC-4P and two 0V9282 mono cameras.

The eeprom is flashed with the stereo calibration (intrinsics + extrinsics)

I am able to get a disparity map by creating a stereoDepth node, but I would like to display a point cloud to get a better idea of what's going on.

I could not get this to work (either with the FFC or with a OAK-D pro). depthai-viewer worked perfect with the OAK-D pro, but returned an error when plugging in the FFC:

stream_enabled=True name='Right' is not configured. Couldn't create stereo pair.

What would be the correct procedure for producing a point cloud from a custom setup ?

Additionnally, we want to try to add the IR projector module, I saw that our ffc needs to be R5 or newer, how can I check that ?

Thanks !

Antoine

apirrone

Ok, regarding the error "Right is not configured" I think it was just a matter of setting all the parameters at calibration (I use my own calibration + flashing procedure these days instead of using luxonis's calibration.py, I must have forgot something.)

Now depthai-viewer displays the depthmap and the pointcloud, but it is extremely bad 🙂

I will try to follow instructions here to improve it

apirrone

Could you give me what you consider to be acceptable calibration results ? I got

right Reprojection Error: 0.516754
left Reprojection Error: 0.473438
Average Epipolar Error is : 0.16123293063130242

Which was good for visual stereo (in VR) but maybe it is not good enough for computing depth maps ?

Thanks !

erik

Hi apirrone ,
Errors seem OK, usually you'd want to have avrg 0.1 or better, but that depth map looks completely off - could you visualize rectified left/right streams?
Regarding the OAK-FFC-IR - you can just check whether your OAK-FFC-4P has this port;

If it does, then it supports the FFC-IR.
Thanks, Erik

apirrone

Hi Erik,

The rectified streams seem ok

apirrone

Ok, I figured it out 😃

The issue was that because we always mount the cameras upside down, I added the ROTATE_180_DEG transformation in my acquisition and calibratino pipeline, but depthai-viewer would compute the stereo rectification on upside down images relative to the calibration data.

The depth and point cloud are much better now, still not perfect but a good baseline to improve upon 🙂