search

tudelft-iv / view-of-delft-dataset

This repository shares the documentation and development kit of the View of Delft automotive dataset.
Other
254 stars 40 forks source link

Alignment of multi-frame radar dynamic object data #96

Closed lotterf closed 5 months ago

lotterf commented 5 months ago

Hello, the data set published by your team is very helpful to me, thank you and your team for your free efforts. However, I still have some questions about the multi-frame radar in the data set. I found that the coordinate system has been calibrated and aligned by the self-driving sensor. However, how should I calibrate the speed of the dynamic objects in the scene? I think there might be some error in the data.

andraspalffy commented 5 months ago

The accumulated multi-frame radar sweeps are indeed transformed to the last position of the sensor. The radial velocities are untouched, i.e. they still give you the radial velocity of the given point e.g. 3 sweeps before towards the sensor position back then (i.e. not now).

lotterf commented 5 months ago

Yes, I know that radial velocity does not change after accumulation, but I would like to know if only radial velocity correction, instead of full velocity correction, will this lead to inaccurate multi-frame radar point cloud information after accumulation? For example, when a dynamic object is moving tangentially in its connection to a radar sensor. At this time, the radial velocity is 0, and there is a certain deviation in the image of the dynamic object between adjacent frames.

andraspalffy commented 5 months ago

I would like to know if only radial velocity correction, instead of full velocity correction, will this lead to inaccurate multi-frame radar point cloud information after accumulation?

Unfortunately I do not understand.