Safety

We  present  two  new  features  for  our  prototype  of  European  Speed  Limit  Support
system:  detection  and  recognition  of  end-of-speed-limit  signs,  as  well  as  a  framework  or detection and recognition of supplementary signs located below main signs and modifying their scope  (particular  lane,  class  of  vehicle,  etc…).  The  end-of-speed-limit  signs  are  globally-recognized  by  a  Multi-Layer  Perceptron  (MLP)  neural  network.  The  supplementary  signs  are detected  by  applying  a  rectangle-detection  in  a  region  below  recognized  speed-limit  signs, followed by a MLP neural network recognition. A common French+German end-of-speed-limit signs recognition has been designed and successfully tested, yielding 82% detection+recognition. Results for detection and recognition of a first kind of supplementary sign (French exit-lane) are already satisfactory (78% correct detection rate), and our framework can easily be extended to handle other types of supplementary signs. To our knowledge, we are the first team presenting results on detection and recognition of supplementary signs below speed signs, which is a crucial feature for a reliable Speed Limit Support.

Mines ParisTech

Valeo Driving Assistance Domain

Presented at the 15th World Congress on Intelligent Transport Systems, November 16-20, 2008, New York, New York

Statistics show that about forty percent of all road accidents occur at urban and rural
intersections, mainly due to misjudgement and rule violation. Collisions involving vulnerable
road users often lead to serious or even fatal injuries. The Intelligent Cooperative Intersection
Safety system (IRIS), as part of the European research project SAFESPOT, is a roadside
application that aims at decreasing the number of accidents at controlled intersections. The
application uses vehicle-to-infrastructure communication to track and analyze the movements
of individual vehicles and explicitly addresses the protection of vulnerable road users. Key
elements of IRIS are the real-time identification of dangerous situations and the application of
appropriate measures to prevent collisions. This paper analyses the safety of vulnerable road
users at controlled intersections within the context of the SAFESPOT-IRIS system.

Peek Traffic bv

Technische Universität München

MAT.Traffic

Presented at the 15th World Congress on Intelligent Transport Systems, November 16-20, 2008, New York, New York

We developed a pedestrian detection system in order to notice a driver the existance of pedestrians. Our previous system which using HOG adaboost classifiers can detect pedestrians with high speed, however, due to the limitation of HOG features, it still has a high false positive rate. In this paper, we propose to cascade the previous pedestrian detector by a false positive remover to improve the accuracy while keeping the high speed. The proposed false positive remover is composed of two adaboost classifiers, one for grayscale image, and another for distance image. The classifier for grayscale image utilize boosted spatial co-occurrence matrix of edge directions --- a kind of improved edge feature proposed in this paper. The classifier for distance image boosts shape features together with spatial co-occurrence matrix of edge directions extracted from the distance image, which is obtained from a stereo camera. Experimental results are presented to validate our method.

Panasonic Corporation

Presented at the 15th World Congress on Intelligent Transport Systems, November 16-20, 2008, New York, New York

any  researches  have  been  conducted  to  materialize  the  driving  safety  support
systems  (DSSS),  including  a  pedestrian  crossing  information  system  and  an  approaching
motorbike  information  system,  aiming  to  prevent  traffic  accidents  at  an  intersection.  A
practical solution to materialize these systems is to use sensors to collect information and for
this  reason various  types of sensors have been proposed. We have been proposing an  image
type  sensor  based  on  the Spatio-Temporal Markov Random Field Model(S-T MRF Model)
used  for  the DSSS.  The  advantage  of  using  the  S-T MRF Model  is  it's  robustness  to  the
occlusion that may happen when some objects such as pedestrians or vehicles moving closely,
resulting a high accuracy of detection. This time we have joined the experiments of the DSSS,
sponsored  by  the UTMS  society  and  developed  the  sensors  used  to  prevent  the  accidents
involving pedestrians or motorbikes.

Omron Corporation

University of Tokyo

Presented at the 15th World Congress on Intelligent Transport Systems, November 16-20, 2008, New York, New York

This paper proposes a method of achieving fast detection of pedestrians, while simultaneously
maintaining good performance regardless of variation in illumination, and in both shape and
scale of pedestrians with a single camera. Regions of interest (ROIs) are acquired by optical
flow fields using the Lucas-Kanade algorithm, and classified by convolutional neural
networks (CNNs) whether they are pedestrians. Detected pedestrians are tracked by using a
particle filter based on adaptive fusion frameworks. The CNNs allow the proposed system to
be robust to variation in illumination and in both shape and scale of pedestrians; and proposed
methods of setting ROIs and tracking pedestrians allow this system to detect a dangerous
situation and warn it to a driver fast. A single camera is only used to conduct this method, thus,
the proposed system is also economically efficient.

Pohang University of Science and Technology (POSTECH)

Presented at the 15th World Congress on Intelligent Transport Systems, November 16-20, 2008, New York, New York