Estimation of the 3D structure of road scenes from monocular images and range data

This paper proposes a method for estimating the 3D structures of road scenes using a monocular camera and a laser scanner. In order to develop advanced driver assistance systems, it is necessary to obtain 3D information concerning obstacles in the road. However, it is generally difficult to obtain accurate 3D structures of road scenes over a wide field of view. Our proposed method estimates 3D structures both from a sequence of images taken by a monocular camera and from range data obtained by a laser scanner. It accurately estimates the camera motion by using both the images and the range data. Moreover, a particle filter is applied to achieve robustness in various scenes. By estimating the camera motion accurately, the 3D positions of feature points that are extracted in the image can be calculated accurately. In our experiments, it has been shown that the proposed method is capable of estimating camera motion in a real road scene and it is able to obtain 3D structure of a road scene accurately over a wide field of view.