search

spoonsso / dannce

MIT License
214 stars 30 forks source link

Could not find enough inliers in imagePoints and worldPoints #136

Closed yuan0821 closed 1 year ago

yuan0821 commented 1 year ago

Hi! does anybody meet the weird error when doing extrinsic calibration? I struggled with extrinsic calibration. The same picture and the same intrinsic parameter; however, whenever I label the 4 points of coordination, the error comes randomly. So I did intrinsic calibration again to ensure the intrinsics were normal(error <0.15 pixels).

Sometimes it works well, and the result of doing Lablel3D is ok; sometimes, everything is failed. Appreciate any suggestions.

image here is my intrinsic matrix: (all camera are the same type, but not sure why camera3 have different parameters) image

yuan0821 commented 1 year ago

Below is the Label3D abnormal result, even though there is no error in the extrinsic calibration. image

data-hound commented 1 year ago

Hi @yuan0821

About the camera3 being different, does it have a different crop factor than the other cameras? Or is the ROI after cropping too far away from the true center of the camera view? Cropping and translation can change intrinsics.

As for the extrinsic calibration failing randomly, that could be due to a variety of reasons. But, keep in mind that under the hood, the estimateWorldCameraPose uses a MSAC algorithm to determine inliers, which is a stochastic algorithm. So, I would expect the output to vary slightly and maybe with only 4 labels a sufficient overlap may not be feasible. I would suggest trying to increase the number of labelled points.