A vanishing point detection method [13]

With increased computing power and without the condition of real-time performance an approach with higher computational effort is reasonable and will be pursued here. As accumulator space we choose the unbounded image plane itself, which avoids a fundamental drawback of most other approaches (see previous section). As accumulator cells we choose (like [11]) the intersection points of all pairs of line segments. We will see that despite the fact that the image plane is unbounded, infinite and finite vanishing points can be treated in the same way.

For the search step we establish three criteria, which vanishing points with mutual, orthogonal directions have to satisfy. The main idea is that orthogonal vanishing points can be used to calibrate a camera. Therefore, one criterion checks the reliability of the deduced camera parameters. On the basis of this, the three vanishing points (accumulator cells) are searched for which have the highest votes (according to the accumulation step) and fulfill these three criteria. This exhaustive search is computationally more expensive than other approaches (like [11,12]).