Authors:
Tan Khoa Mai
1
;
Michèle Gouiffès
2
and
Samia Bouchafa
1
Affiliations:
1
Univ. d’Evry Val d’Essonne and Université Paris Saclay, France
;
2
Univ. Paris Sud and Université Paris Saclay, France
Keyword(s):
Optical Flow, C-velocity, 3-D Plane Reconstruction.
Related
Ontology
Subjects/Areas/Topics:
Image Processing
;
Informatics in Control, Automation and Robotics
;
Intelligent Transportation Technologies and Systems
;
Robotics and Automation
;
Vision, Recognition and Reconstruction
Abstract:
This paper deals with a new method that exploits optical flow field properties to simplify and strengthen the
original c-velocity approach (Bouchafa and Zavidovique, 2012). C-velocity is a cumulative method based
on a Hough-like transform adapted to velocities that allows 3D structure identification. In case of moving
cameras, the 3D scene is assumed to be composed by a set of 3D planes that could be categorized in 3 main
models: horizontal, lateral and vertical planes. We prove in this paper that, by using directly pixel coordinates
to create what we will call the uv-velocity space, it is possible to detect 3D planes efficiently. We conduct our
experiments on the KITTI optical flow dataset (Menze and Geiger, 2015) to prove our new concept besides
the effectiveness of uv-velocity in detecting planes. In addition, we show how our approach could be applied
to detect free navigation area (road), urban structures like buildings and obstacles from a moving camera in
the context o
f Advanced Driver Assistance Systems.
(More)