Providing intrinsics for extrinsic calibration with ftheta model

Autonomous Vehicles DRIVE AGX Orin DRIVE AGX Orin General
1

Specs:
Platform: DRIVE AGX Xavier
Driveworks: 5.1.0
DriveOS: 6.0.6

Hey there, I wanted to use the calibration-graph-cli tool and provide the intrinsics with the ftheta model. For the intrinisc calibration I am using a different tool which provides the following output. Unfortunately I am not able to convert this into the intrinscs/<camera>.json file in such a way that it works.
This are the intrinsic parameter I get:

width: 1936 
height: 1216
intrinsics:
  cx: 966.461471
  cy: 602.120647
  fx: 981.817474
  fy: 979.962902
  distortion_model: f-theta
  k:
  - -0.0394357 [radial distortion coeff 1]
  - -0.00190517 [radial distortion coeff 2]
  - -0.0277921 [tangential distortion coeff 1]
  - 0.0173957 [tangential distortion coeff 2]

This is how I would convert it regarding to this thread:

{
    "constraints": [],
    "focalLengthPixel": [
        981.817474,
        979.962902
    ],
    "image-size": [
        1936,
        1216
    ],
    "model": {
        "bw-poly": [
            0.0, [0.0 as written in the paper]
            0,001018519, [1/fx as written in the paper]
            -0.0394357,
            -0.00190517,
            -0.0277921
            0.0173957
        ],
        "center": [
            966.461471,
            602.120647
        ],
        "fov": 1.5566234691909453,
        "image-size": [
            1936,
            1216
        ],
        "is-model-fixed": true,
        "name": "back",
        "type": "ftheta"
    },
    "name": "back",
    "type": "ftheta"
}

I also tried running the tool with different --ftheta-poly-degree=4/5 --ftheta-reference-poly=0/1 but without getting the expected result.

3

Dear @jan73,
Could you add more details on what is the issue? Do you want guidance on parameter mapping or tool usage?

4

If I use this I get the following error message:

Terminating app due to unexpected exception:
DW_INVALID_ARGUMENT: FThetaCamera: First poly coefficient must always be zero

In general a documentation on the intrinsic.json file format would be great. In terms of which key does affect what in case you want to build you own intrinsic file. But also an explanation would be great on how to remap the parameters from a common ftheta model to work with the nvidia tool since it does not simply expect: distortion model: equidistant distortion_coeffs: [k1 k2 k3 k4] as given whne using openCV.

5

Dear @jan73,
May I know why you are using other tool for intrinsic calibration? Are you trying to get similar to DriveWorks SDK Reference: Intrinsics Constraints Tool tool?

6

Hey @SivaRamaKrishnaNV, we have several other use cases where we need the intrinsics and since we don’t know how to convert between the output from our tool and the output from the Nvidia tool we cannot use the Nvidia tool.