Abstract
We propose a method to estimate camera motion and road width (distance to buildings in both ides) from videos taken in urban environments. This facilitates applications like localization by educing the search space. The camera motion between frames is estimated. We propose that rotation nd translation are independently estimated. The advantage of this method is more robust and faster rocessing. A plane is fitted to the wall in each side, and the distance is estimated using the quation of a homography induced by a plane. The experiments showed this method could iscriminate different kinds of roads successfully.[This abstract is not included in the PDF]
