Omnistereo with the omnipolar camera
This section presents supplemental videos related to our omnipolar camera (to be published at ICCP 2013). For other supplemental material related to previous work on omnistereo (ICCV 2011, TAP2012), please click here.
We here present the Lab scene captured using an omnipolar setup using 3 cameras with fisheye lens lying on a circle with a diameter of about 12cm. The camera setup was fixed on a rail on the ground to produce a travelling of about 1.5m while a person is walking by.
The next output video is formatted in anaglyph for a cylindrical display. The full 360 degree cylindrical images are shown in three 120 degree parts (these parts do not correspond to the 120 degree field of view used in each camera for the left or right eye).
Download: MP4
Monocular view:
Download: MP4
The next output video is formatted in anaglyph for a dome. Vertical parallax was adjusted so that the ceiling above the cameras was correctly aligned.
Download: MP4
Omnistereo by rotating 2 cameras on a tripod
This section contains supplemental videos related to our ICCV 2011 publication titled Panoramic Stereo Video Textures. Video compression can slightly reduce the effectiveness of the stereo effects for the anaglyph sequences. Also, the effectiveness depends on the amount of (typcally red) leakage through one’s red/cyan glasses.
Download: MP4
The left camera component of the rendered stereo motion texture panorama. Here we show the 360 degree panorama as 3 panels of 120 degrees each. Here we do not blend adjacent blocks as in Figure 12a which allows the reviewers (1) to examine the underlying motion discontinuities at the block boundaries and (2) to appreciate that there are significant exposure and/or lighting variations between adjacent blocks. In the original captured video sequence, the camera is rotating clockwise i.e. left to right, which corresponds to the right drifting block boundaries. The rendered sequence loops 5 times.
Download: MP4
The stereo motion texture panorama of the same river sequence as in the previous example. However, rather than showing panels that simultaneously cover 360 degrees, we instead show a higher resolution frame, namely we show a synthesized pan which simulates the camera rotating in the opposite direction, i.e. counter-clockwise. This video was obtained by cropping the computed 360 dgree panorama to a different window location in each frame. The block boundaries are not visible because blending was used, as in Figure 12b. As described in the paper, the minimum overlap between adjacent blocks was 20%.
Download: MP4
The same description as the previous sequence applies, but now we show the field sequence from Figure 1 in the paper.
Download: MP4
Here we show the result of our technique on the (monocular only) Yachts sequence, which is taken from the paper ‘Panoramic Video Texture’ by Agarwala et al (2005).
Perception Experiments
To better understand the visibility of ghosting and how it might vary from scene to scene, we carried out a perceptual experiment (TAP 2012). The experiment measured how fast naive observers could detect moving blended regions for different blending widths and for different scenes. Below, you can download monocular versions of the stimuli videos for the 5 scenes we used. Although blending location was varied during the experiments, the blended region is here in the left part of all videos.
|
Flowers (large blend) Download: MP4 |
Flowers (small blend) Download: MP4 |
|
Bush (large blend) Download: MP4 |
Bush (small blend) Download: MP4 |
|
Smoke (large blend) Download: MP4 |
Smoke (small blend) Download: MP4 |
|
Lake (large blend) Download: MP4 |
Lake (small blend) Download: MP4 |
|
River (large blend) Download: MP4 |
River (small blend) Download: MP4 |