Road Region Detection by Spatio-Temporal Graph Segmentation of Optical Flows using On-Board Camera

Kenji Nishida, Jun Fujiki, Takumi Kobayashi, Chikao Tsuchiya, Shinya Tanaka, and Takio Kurita

Keywords

Road Detection, ITS

Abstract

In this paper, we present a novel algorithm for road region detection using an on-board camera. The algorithm first detects optical flows by matching features of image patches that have sufficient feature points. Then, the physical model of optical flows for the entire road region is estimated from the detected optical flows using RANSAC. The validity of the estimated flows is verified from the flow model, and the road region is segmented accordingly by using the graph-cut technique for each video frame. Finally, the road region is further refined by voting with three consecutive video frames. The experimental results indicate that the proposed algorithm is effective in detecting the road region in suburban traffic.

Important Links:

Go Back