Road detection for inner-city scenarios remains a
difficult problem due to the high complexity in scene layout
with unmarked or weakly marked roads and poor lighting
conditions. This paper introduces a novel method based on
multi normalized-histogram with Joint Boosting algorithm to
road recognition. The approach performs three modules in
parallel that are the Image Segmentation, the Texton Maps and
the new one Dispton Maps. The first one applies a combination
of pre-filters with Watershed Transform to make the superpixel.
The last two perform a dense feature extraction based
on 2D texture image and 3D disparity image to get appearance,
shape and context information. At last, a discriminative model
of road class is learned based on distribution of Textons and
Disptons applied in Joint Boosting algorithm. The proposed
work reports real experiments carried out in a challenging
urban environment utilizing the modern KITTI benchmark
for road areas in which meaningful evaluation can be done
to illustrate the validity and application of this approach.
Index Terms— Road Detection, Computer Vision, Joint
Boosting, Texton Map, Dispton Map, Watershed Transform