01697345

Component

Collecting road curvature data is crucial for both Road Asset Management Systems (RAMS) and Advanced Driver Assistance Systems (ADAS). This paper proposed a novel vision-based approach for horizontal road curvature computation. The authors first demonstrate that sufficient central angle is crucial for robust and accurate circle fitting from an arc segment. To achieve this goal, they adaptively use a number of sequential images for curvature computation to meet sufficient central angle requirement. And the authors proposed a modified Iterative Closest Point (ICP) method to adaptively register curves reconstructed from multiple video log images. Especially, the initial guess for ICP is derived from point correspondences by utilizing epipolar geometry constraint to ensure fast speed and avoid local minimal. Second, an error analysis model is applied to filter those reconstructed points with large reconstruction errors to further enhance curvature computation. Finally, road curvature is computed by fitting circle from the adaptively registered curves. The proposed method has been tested with the actual video log images collected on different curved road segments with the radii of 115m, 275m, and 430m, respectively. The results demonstrate that the proposed method is practical, reliable, and accurate to compute different degrees of road curvatures.

This paper was sponsored by TRB committee ABJ60 Standing Committee on Geographic Information Science and Applications.

19-02371

English

17p

2019

Transportation Research Board 98th Annual Meeting

Digital/other

Figures; Photos; References; Tables

Algorithms; Asset management; Driver support systems; Error analysis; Highway curves

Video logging

Administration and Management; Data and Information Technology; Highways

Transportation Research Board Annual Meeting 2019 Paper #19-02371

TRIS, TRB, ATRI

Dec 7 2018 9:24AM