Minimal Aspect Distortion (MAD) Mosaicing of Long Scenes
Alex Rav-Acha, Giora Engel, Shmuel Peleg
Abstract
Long scenes can be imaged by far away cameras, or by mosaicing multiple images from closer cameras scanning the scene. We address the case of a video camera scanning a scene while moving in a long path, e.g. scanning a city street from a driving car, or scanning a terrain from a low flying aircraft. A new minimal aspect distortion (MAD) mosaicing of the long scene is presented, which uses depth to minimize the geometrical distortions and to improve stitching of long panoramic images. We also present a robust "direct" method to provide both camera motion and dense depth. A limited motion model of a scanning camera is used to increase robustness of motion computation. Iterative graph-cuts, with planar labels and a flexible similarity measure, is used for depth computation. Interactive visualization using X-Slits is demonstrated given the computed motion and depth. In this case the depth information is used for better stitching of the mosaic. The presented approach is very robust, and works on sequences having thousands of frames even when using a hand-held camera. Examples are shown on a few challenging sequences. |
Technical Report describing this work (pdf)
Demos
(Click on pictures to enlarge or to view the video)
A Street in Jerusalem
Boat Ride in Germany
Derailed Shinkansen Train