AN ANALYSIS OF CROWDS FLOW CHARACTERISTICS 
BY USING LASER RANGE SCANNERS 
K. Katabira 3 , T. Suzuki 3 , H. Zhao b , Y. Nakagawa c , R. Shibasaki 3 
a Center for Spatial Information Science, University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa-shi, Chiba, Japan - 
(katabira, t-suzuki, shiba)@csis.u-tokyo.ac.jp 
b School of Electronics Engineering and Computer Science, Peking University, Beijing 100871, China - 
zhaohj@cis.pku.edu.cn 
c Research and Development Center of JR East Group, East Japan Railway Company, 2-0 Nisshin-cho, Kita-ku, 
Saitama-Shi, Saitama - nakagawa@jreast.co.jp 
KEY WORDS: Laser scanning, Sensor, Computer Vision, Feature extraction, Spatial Planning, Machine vision 
ABSTRACT: 
Recently, according to growth and concentration of population in urban area, congested situations at several railway stations in rush 
hour become a serious social issue. Therefore, designing the station more comfortably for the passengers has been desired. In other 
words, in order to mitigate the congested situations, a method measuring crowds flow characteristics should be developed. In this 
paper, we propose a people tracking method for wide and high density situations by using multiple laser range scanners. 
Additionally, we try to analyze the crowds flow characteristics such as traffic volume, traffic density, degree of intercrossing, 
velocity, and directionality. Then, we detect several prominently congested areas. In order to evaluate our proposed system, an 
experiment at Japan railway station was conducted. The accuracy of people tracking shows over 80% in rush hour. Based on the 
obtained trajectories, the density and direction distribution of the trajectories are calculated by using Kernel Density Estimation 
method. Furthermore, based on the variance of moving direction and traffic volume, we define a degree of intercrossing in each flow, 
and then the prominently congested areas are highlighted. Finally, we conclude our proposed system will be efficient to mitigate a 
congested situation and improve the usability of the public space such as railway station, exhibition hall, or shopping mall. 
1. INTRODUCTION 
In public space such as railway station, museum, exhibition hall 
or shopping mall, to measure people behaviour or flows 
becomes increasingly important for several reasons, e.g., floor 
planning, mitigation of congestion, detection of suspicious 
person for security, alignment of advertising messages, etc. 
Especially, at the concourse of railway station, congested 
situation in rush hour becomes a serious social issue. Therefore, 
an automatic people tracking system has been desired, which 
achieves with high accuracy even if the target space is wide and 
congested. Conventional researches trying to track multiple 
pedestrians had almost used CCD camera (Haritaoglu, 2000; T. 
Zhao, 2004; Curio, 2000). However, camera has some 
disadvantages, which are affected by illuminating disturbance, 
narrow viewing angle, and a problem of personal privacy. 
Therefore, we try to exploit multiple laser range scanners. 
Moreover, in order to extract the characteristics of pedestrians’ 
behaviour or people flow, several methodologies had proposed, 
such as Semantic Scene Model (Makris, 2005), Acitivity Map 
(Demirdjian, 2002), and other algorithms (Hu, 2006; Brandle, 
2006; Grimson, 1998; Stauffer, 2000). However, conventional 
researches had targeted at few persons and relatively narrow 
space, so that almost researches have not dealt with much 
congested situation e.g. a railway station. In this paper, we 
propose a method to track pedestrians robustly for a congested 
situation and to analyse crowds flow characteristics such as 
traffic density, walking direction, walking velocity, a degree of 
intercrossing, and those relativity. Finally, we describe an 
experimental result conducted at concourse of railway station in 
Japan. 
2. METHODOLOGY 
2.1 Outline of Laser Range Scanner 
In order to track multiple pedestrians, single-row type laser 
range scanner, LMS200, by SICK corp. is exploited. It 
measures a range value between sensor and surrounding objects 
by calculating a time of flight of a reflected laser pulse, and it 
scans horizontal plane of 180 degrees by rotating an inner 
mirror. The laser scanner has high angle resolution (0.5 degree) 
and high scanning rate (37.5Hz). It uses an eye-safe laser beam 
(safety class 1) with 905nm wavelength. The laser beam covers 
a maximum range of 30m, and the range error is within 4cm. 
The advantages of laser range scanner are to obtain range value 
actively (light is not necessary), and it has wide field of view. 
In this paper, in order to cover a wider area and mitigate 
occlusions, multiple laser scanners are installed on a floor at 
20cm height, and scan horizontally to aim at pedestrians’ feet as 
shown Figure 1, 2. Therefore, obtained data consist with same 
horizontal level. Each laser scanner is controlled by a client 
computer, and the client computers are connected to a server 
one by using LAN. The obtained range data are gathered to the 
server through the network on the real time basis. 
Figure 1. Sensor alignment scanning pedestrians’ feet 
953