Objects recognition in traffic scenes by using multiple laser range scanners
Abstract
In this research, we propose a method for recognizing moving objects in railway crossing such
as pedestrians, bicycles, and cars, using a number of single-row type laser range scanners in
order to prevent traffic accidents. Although pedestrians and bicycles may get involved in the
accidents, current obstacle-detectors recognize only large objects like cars and trucks, because
railway companies have paid more attention to detect larger obstacles that can cause a crucial
damage for train. In recent years, however, tragic accidents of the elderly people in a wheelchair
at the railway crossing attract more public attention. It becomes more important to analyze and
monitor traffic objects more effectively. We apply laser range scanners with high scanning ratio,
wide viewing angle and long-range measurement capability to the traffic object monitoring.
An experiment was conducted by using four laser range scanners to monitor the traffic objects in
railway crossing. Each laser scanner is located in different places, scanning at a horizontal plane
about 16cm above the ground, which is controlled by a client computer. All client computers are
connected through local area network (LAN) to a server computer, which gathers the laser
points of moving objects from all client computers, and conducts tracking process. Server
computer gathers the data of moving objects in the latest range frames from all client computers
and integrates them into a global coordinate system. Since multiple laser beams hit an identical
object because of high angle-resolution resulting, laser points are clustered based on the relative
distances and specific shapes of moving objects.
1.Introduction
In the recent years, it is demanded to recognize various moving objects in railway crossing such
as pedestrians, bicycles, and cars, in order to prevent traffic accidents. Current
obstacle-detectors recognize only large objects like cars and trucks, because railway companies
have paid more attention to detect larger obstacles that can cause a crucial damage for train.
However, they must pay attention to pedestrians and bicycles not to get involved in the accidents.
With the arrival of aging society and spread of barrier-free services, elderly people and
physically disabled people will have more occasions to go out and to cross railway crossing.
Furthermore, if the differences can be recognized between large objects (like cars and trucks)
and small objects (like pedestrians and bicycles), railway companies can avoid unnecessary
emergency stops by just making announcement or warning of security assurance for pedestrians.
In the case of experiment that aims to track specific objects, we can easily acquire trajectory
data by providing objects with positioning sensors such as GPS. However, in the case of
tracking the unlimited number of moving objects in railway crossing, it’s difficult to obtain the
trajectory data, because we cannot provide them all with equipment. In addition, we must detect
moving objects in all part of railway crossing.
Video based approaches have some disadvantages, such as narrow field of view, coverage
limitation of setting, and susceptibility to change in light condition. In this research, we use
multiple laser scanners to resolve above problems. This sensor has several advantages of high
scanning rate, wide viewing angle and long-range distance. On the other hand, laser scanner
cannot detect transparent, very low reflectance objects (e.g. dark color (blue, black)), and high
reflectance objects that can cause strong specular reflection. Therefore we also need to
investigate more appropriate measurement methods.
