Superpixel-based Road Segmentation for Real-time Systems using CNN
Farnoush Zohourian
1
, Borislav Antic
2
, Jan Siegemund
2
, Mirko Meuter
2
and Josef Pauli
1
1
University of Duisburg-Essen, Department of Computer Science and Applied Cognitive Science,
47048, Duisburg, Germany
2
Delphi Electronics & Saftey, D-42119, Wuppertal, Germany
Keywords: Superpixel, Semantic Segmentation, CNN, Deep Learning, Road Segmentation.
Abstract: Convolutional Neural Networks (CNN) contributed considerable improvements for image segmentation tasks
in the field of computer vision. Despite their success, an inherent challenge is the trade-off between accuracy
and computational cost. The high computational efforts for large networks operating on the image’s pixel grid
makes them ineligible for many real time applications such as various Advanced Driver Assistance Systems
(ADAS). In this work, we propose a novel CNN approach, based on the combination of super-pixels and high
dimensional feature channels applied for road segmentation. The core idea is to reduce the computational
complexity by segmenting the image into homogeneous regions (superpixels) and feed image descriptors
extracted from these regions into a CNN rather than working on the pixel grid directly. To enable the necessary
convolutional operations on the irregular arranged superpixels, we introduce a lattice projection scheme as
part of the superpixel creation method, which composes neighbourhood relations and forces the topology to
stay fixed during the segmentation process. Reducing the input to the superpixel domain allows the CNN’s
structure to stay small and efficient to compute while keeping the advantage of convolutional layers. The
method is generic and can be easily generalized for segmentation tasks other than road segmentation.
1 INTRODUCTION
One of the long-lasting goals of computer vision is the
automated scene understanding from a variety of ima-
ges. Exposing image specification is useful for appli-
cations, like image editing, image search and environ-
ment perception for autonomous vehicles. Detecting
objects like roads, pedestrians, vehicles, traffic signs,
etc. is important for many driver-less cars and driver
assistance systems. Due to the variability of different
factors like colour, shape, illumination and shadows
or obstacles on the road surface, the road detection is
a challenging problem. The state of arts techniques to
solve this problem are mainly based on deep learning
and Convolutional Neural Networks (CNNs) (LeCun
et al., 2015; Schmidhuber, 2015). These methods ena-
ble towards better visual understanding by applying a
semantic segmentation process in which each pixel is
assigned to an object category. The segmentation re-
sult provides meaningful information to support hig-
her level scene understanding tasks.
Currently, there are two major approaches to train
CNN-based image processing systems. The two ap-
proaches differ with respect to the input data mo-
del. One of the approaches is based on a patch-wise
analysis of the images, i.e. an extraction and clas-
sification of rectangular regions having a fixed size
for every single image (Ciresan et al., 2012; Farabet
et al., 2013; Ganin and Lempitsky, 2014; Ning et al.,
2005). The other one is based on full image reso-
lution, wherein all pixels of an image in the original
size are analyzed (Long et al., 2015). Most recent
improvements in both CNN-based methods were ac-
complished by increasing the network size (Simonyan
and Zisserman, 2014; He et al., 2016), whereas dee-
per networks provoke large computational costs that
make them unsuitable for embedded devices in driver
assistance systems.
In the current work we apply a superpixel-based
CNN method for the specific application of pixel-wise
road segmentation that uses superpixels as input data
model. To the best of our knowledge, it is the first
time that irregular superpixels with regular lattice pro-
jection for Convolutional purpose is given as input
data model into a CNN network. The proposed met-
hod comprises the following steps: first, segmenting
the image into superpixels, wherein the superpixels
are coherent image regions comprising a plurality of
Zohourian, F., Antic, B., Siegemund, J., Meuter, M. and Pauli, J.
Superpixel-based Road Segmentation for Real-time Systems using CNN.
DOI: 10.5220/0006612002570265
In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 5: VISAPP, pages
257-265
ISBN: 978-989-758-290-5
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
257