Calibration not working? about pie HOT 5 CLOSED

Hi Luke,

Thanks for using our dataset. We did our best to generate calibration data as best as we can. Unfortunately, so far we have used the information in some experimental code. As a result, I cannot make any definitive comments on how well it works.
On a different note, however, based on the image you posted, I don't think the 5 meters mark is too off. One way to confirm that can be to measure the size of a known object in the scene and calculate the distance based on its actual size.

Amir

from pie.

Hi Amir,

thank you for a prompt reply!

Could you maybe comment about which elements of the calibration data are definitely correct and which you are perhaps not as sure about? 😉

For example, it seems to me that the intrinsics (K matrix) seem reasonable, but the extrinsics = camera rotation is probably a bit more than 10 degrees, and there also seem to be some rotation around the X axis. And not sure about the distortion coefficients?

Thank you
Lukas

from pie.

Hi Lukas,

Once again, I cannot give a definitive answer. I guess the external parameters are more prone to error given that we used the dashcam to record. We kept the position of the camera fixed all the time and used proper tools to measure angles. However, there is a possibility that the position of the camera be altered a bit because of some bump or a whole on the road. I don't, however, think that is significant.

from pie.

@lukeN86 Hi Lukas,
I compute the intrinsics matrix using your script and the results is [[1065.29876, 0.0, 959.5], [0.0, 1064.82575, 539.5], [0.0, 0.0, 1.0]]. Have you test this? Thank you.

from pie.

Closing due to inactivity

from pie.