TRACKING PEDESTRIAN BY USING MULTIPLE LASER RANGE SCANNERS
Katsuyuki Nakamura“, Huijing Zhao“, Ryosuke Shibasaki“
Kiyoshi Sakamoto”, Tomowo Ooga“, Naoki Suzukawa“
a Center for Spatial Information Science, The University of Tokyo, JAPAN
^ Frontier Service Development Laboratory, East Japan Railway Company, JAPAN
* Asia Air Survey Co.,Ltd, JAPAN — ^ JR East Consultants Company, JAPAN
katsuQiis.u-tokyo.ac.jp
Commision IV, WG IV/1
KEY WORDS: Laser scanning, Tracking, Detection, Recognition, Acquisition, Systems, Application, Video
ABSTRACT:
Tracking pedestrians is
are
system of tracking
are located in different places,
model is defined, based on which
used about 250,000 passengers a day. Pedestrians’ trajectories are extracted and the extracted results are comp
an important topic in the field of pattern recognition and image understanding due to its various applications.
Although video-based approaches have been studied for decades, restricted setting location, narrow viewing angle and limited resolution
the major obstacles to achieve higher accuracy and to apply to a wider variety of applications. In this paper, we propose a nove
pedestrians in a wide and open area using a network of single-row type laser range scanners. Laser range scanners
scanning pedestrians’ feet at a horizontal plane about 16cm above the ground. A pedestrians’ walking
a tracking algorithm is developed. An experiment is conducted in a railway station in Japan, which is
]
ared with real situations
that are recorded using video cameras. It is concluded that this system has very high effectiveness.
1 INTRODUCTION
Human positioning data have been widely used in various fields
such as architecture, disaster prevention and traffic engineering.
Because of suchlike various applications, there are a lot of re-
searches to measure pedestrians’ trajectory. Great majority of
these works focus on motion analysis through video image, e.g.
Chia-Jung, 2004; Rosales, 2003; Ohta, 2003. On the other hands,
video-based approaches have some disadvantages, such as nar-
row FOV (Field of View), limited resolution. Multiple cameras
are always needed to cover a large area, to reduce occlusion and
to solve crossing problem as well. However, it is practically diffi-
cult to digitally fuse the data of multiple cameras, which requires
accurate calibration and complicated calculation between differ-
ent perspective coordinate system. Therefore, only few systems
have been applied to the measurement of high-density crowds in
wide area, such as that of railway stations and exhibition halls.
In this research, we propose a novel system for tracking pedes-
trian over a wide and open area using a network of single-row
type laser range scanners (briefly called laser scanner in the fol-
lowings). We apply this system to measure passengers' flow in a
railway station. The results are compared with real situations that
are recorded using video cameras.
2 METHODS
In this section, we first briefly introduce the sensor system for
pedestrians' tracking. A description to the tracking algorithm is
given next, followed by an address to efficiency and accuracy as-
sessment to the system.
Figure 1: Pedestrian Measurement by using Laser Scanner
2.1 Sensor
Single-row type laser range scanners produced by SICK corp. are
employed. It measures range distances from the sensor to sur-
rounding objects using the method of time-of-flight. Laser scan-
ner has the advantages of direct measurement, high accuracy (av-
erage distance error is 4cm), wide viewing angle (180 degree) and
long range-distance (maximum range distance of 30m). More-
over, it has high angle-resolution of 0.5 degree because of a little
diffusion of laser beam. Frequency is 37.5 Hz and wave length of
laser beam is 905 nm (near-infrared).
In this research, scanners are set on the floor for horizontal scan-
ning at an elevation of about 16.3cm above the ground. Cross sec-
tional data at the same horizontal level containing both moving
and static objects are obtained in a rectangular coordinate system
of real dimension. Figure 1 shows a laser scanner in experimental
site.
In addition, we use multiple laser scanners to cover a large area,
and to reduce occlusion as well. An integration of each scanner
is conducted by using Hermart transformation that deals with a
shift and a rotation. Transformation-parameters are calculated by
1260
Internati
matchin;
poles by
these pr
easily ca
on regist
Shibasak
2.2. Pet
The flow
parts: 1)
Pedestria
cesses in
tion on fc
2002.
2:34 :: B
range sca
all range
threshold
measured
backgrou
angles. A
images tc
objects.
222 P
processes
grouping
date.
Multiple
resolution
grouped t
number o:
grouping |
the 30cm
conducted
in previou
from the
trajectorie
2.2.3 ; Pe
forward, c
foot swing
roles by I:
forward. ,
Mochon a
initial pos
the swing
the swing
Here initiz
meets toge
celeration,
tively, whe
to a horizo
coordinate
In the cas
right foot.
shifts from
at an accel
