Geometric and Statistic Constraints in Dynamic Re-Orientation of On-Board Camera: Inherent Vanishing Points

This paper investigates the stochastic projective properties of natural line segments in highway scenes for the purpose of dynamically re-orientating the on-board camera. We demonstrate there are geometric and statistic constraints on the projection of such line segments. A new concept of Inherent Vanishing Points associated with land vehicle boards was employed to relate to the observed line segments and rotation matrix. We show how these constraints greatly simplify the re-orientation of the on-board camera. We validate a dynamic re-orientation algorithm based on these constraints with a real image sequence. Compared with the techniques based on static calibration fields, the method based on our proposed statistical constraints is effective, stable, and easy to use.