Vision-based road region detection using probability map of color features of road
Jae-Hyun Lim, Junho Chung, Tae‐Koo Kang, Myo–Taeg Lim · 2017
This paper proposes a road region detection algorithm using a vision sensor. The proposed algorithm uses color features of road to detect the road region. We first segment the image into patches to reduce the computational complexity and the noise. Superpixel method is applied to the segmentation instead of square patches for the precise segmentation. After the segmentation, similarities of patches are calculated by undirected graph-based shortest path algorithm. The distance between neighboring patches is defined as the Euclidian distance of CIE-Lab color and Illumination-invariant color. By using the similarity of patches, isolated region of the image and similarity of patches with bottom of the images are obtained. To ensure robust performance even when the non-road region is located at the bottom of the image, a probability map is constructed by combining isolated region of the image and similarity of patches with bottom of the image. Experimental results show the robustness of the proposed algorithm in various conditions.