Image stitching using gradual image warping in autonomous driving

To improve object recognition and tracking in au-
tonomous driving, we first create a seamless panorama. Object
recognition can benefit from image stitching, especially at the
borders of individual images when an object is only partially
visible. This also prevents duplicate detection of the same ob-
jects in overlapping image areas that are to be filtered for track-
ing. In this process, a homography is determined for the over-
lapping image area, whereby the entire image is transformed
using classical image stitching methods. As a result, the defor-
mations propagate to further images that are to be added to the
panorama. To avoid this problem, we integrated a step-by-step
image warping approach into our existing stitching pipeline.
This ensures that after attaching one image to another, the out-
ermost right and left borders of the panorama are no longer de-
formed. Furthermore, the panorama width remains constant re-
gardless of the calculated homography. We have evaluated our
approach on the nuScenes dataset and the Waymo Open Dataset
for perception. In addition to a qualitative assessment, we eval-
uate the resulting panoramas in terms of the deformation of the
... mehr
individual images as well as the deformation of labeled object
instances.