In this thesis, we propose a scheme for determining the perspective structures of the corridor for intelligent mobile robot navigation systems. The proposed scheme is derived from the 3-D geometrical constraints such that the given scene of the corridor is projected onto the image plane by the perspective projection, and thus most of the parallel lines in real scenes meet at least one point in the image plane called the vanishing point. Since the vanishing point gives the important clues in determining the perspective structures of the corridor, we design a scheme that can reliably determine those vanishing points from monocular images. First, we extract the line segments from the input images and then map those line segments onto the Gaussian Sphere. In the Gaussian Sphere, a line segment in 2D image is represented as a circle rounding the sphere and the point at which any two circles meet directly corresponse to the vanishing point of the two lines in the 2D image plane. Thus it can be easily determined the major vanishing points of the corridor by searching the maximal points on the Gaussian Sphere. Finally, based on the extracted vanishing points, we modify the 3-D model of the corridor which has been constructed by the 3-D geometrical constraints and determine the perspective structures of the corridor. We exploit the designed scheme to determine the heading directions of the intelligent mobile robot in real environment and show some experimental results.
