Closed nuyhead-hwang closed 1 year ago
First of all, the camera projection module should be changed to Hilti's distortion model Equistant. Besides, the processing of lens distortion is overloaded in vikit library, i.e., world2cam
and cam2world
. As for the points that are not visible in the image, we will not use them in the VIO submodule.
Hello? Could you please check this issue again ?
@xuankuzcr
I also tried to test the HILTI 2021 dataset.
But trajectory which is estimated by FAST-LIVO very unstable and eventually diverged.
I set the parameters provided here as shown in the code below. but FAST-LIVO diverge during the dataset playback.
And there is no part where the lens distortion is processed in the code. why is that ?
And when the 3D point cloud of the rotating LiDAR is re-projected onto the image, how to deal with rear area pointclouds that are not visible in the image ?
Originally posted by @vislero in https://github.com/hku-mars/FAST-LIVO/issues/37#issuecomment-1436348480