Here you can find some of the spherical imaging related stuff we (me and my collegue Benjamin Huhle) recorded over the last years at WSI/GRIS. Hopefully, it is useful for experimenting with this special kind of image data. The virtual images have been rendered using RadioLab and the real scenes were acquired using a LadyBug2 spherical camera system. All omnidirectional images are in the equirectangular or latitude/longitude format, most of them are 8-bit JPGs or BMPs.

Rotation Estimation

Virtual Image Sequences

Four sequences of virtual omnidirectional images of size 1024x512 pixels rendered in RadioLab. Assuming a right-handed coordinate system, the center of the images corresponds to the negative Y axis and the Z axis coincides with the vector from south to north pole.

  • Rotation around X axis from [-90°,90°], 181 frames: download (53MB)
  • Rotation around Y axis from [-90°,90°], 181 frames: download (52MB)
  • Rotation around Z axis from [-90°,90°], 181 frames: download (48MB)
  • Rotation around XZ axis from [-90°,90°], 181 frames: download (53MB)

Real Image Sequences

Three sequences of omnidirectional images of size 512x256 pixels acquired with a Ladybug2 camera mounted on a highly-accurate pan-tilt unit. The camera system consists of a calibrated set of six wide angle cameras covering about 360°x130° of the entire sphere and performes the stitching automatically. Note, that the camera origin does not coincide with the rotation axis. Therefore, an additional translational movement (up to ≈20cm and ≈50cm) is induced. Also, the camera is not aligned with the rotation axes!

  • Masked area of the camera image in latitude/longitude representation: download
  • Rotation around pitch axis from [-90°,90°], 180 frames: download (45MB)
  • Rotation around roll axis from [-65°,65°], 130 frames: download (33MB)
  • Combined rotation around roll axis from [-65°,65°] and pitch axis from [-90°,90°], 180 frames: download (47MB)

Template Tracking

Real Image Sequence

In this sequence, the camera is rotated arbitrarily while being attached to a tripod. Again, there's a slight translational offset of the camera's optical center and the rotation center and there is no ground-truth orientation information. We used this scene to perform tracking of a small template.

  • Masked area of the camera image in latitude/longitude representation: download
  • Mask used for template tracking: download
  • Sequence of omnidirectional images (300 frames): download (17MB)
  • Sequence of omnidirectional images (939 frames) with high framerate that contains large translational movements and close objects: download (123MB)