Efficient plane detection from a single moving camera

This paper deals with the detection of any kind of parameterized surface without camera calibration or a priori knowledge about the vehicle egomotion. We assume in this first study that the 3D scene is a set of 3D planes that we propose to detect exploiting iso-velocity curves bound to an estimated optical flow. We believe that cumulative approaches based on a voting scheme are very efficient in making more robust the detection and coping with optical flow imprecision. Therefore, we design a Hough-like frame called c-velocity. We show how this c-velocity could serve exhibiting any plane. 3D-planar structures like obstacles, road or buildings are focused on in the context of a driver assistance application. Under a translational camera motion, planar surfaces are transformed into parabolic of the c-velocity space.