DEVELOPMENT OF TECHNIQUES FOR VEHICLE MANOEUVRES RECOGNITION 
WITH SEQUENTIAL IMAGES FROM HIGH ALTITUDE PLATFORMS 
T. Fuse*, E. Shimizu*, R. Maeda® 
* Dept. of Civil Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan - 
: (fuse, shimizu)@planner.t.u-tokyo.ac jp 
Tokyo Gas Corporation, 1-5-20 Kaigan, Minato-ku, Tokyo, Japan 
Commission V, Working Group IC V/III 
KEY WORDS: Tracking, Monitoring, Image Sequences Analysis, Pattern Recognition, Planning, Navigation 
ABSTRACT: 
Traffic flow surveillance for traffic control and transportation planning require fixed-point continuous observation of exact dynamic 
vehicle movement. High altitude platforms are expected to resul 
continuous observation. These high resolution and continuous im 
result, the high altitude platforms have a great potential with wide 
trajectories. In this study, we explore the possibility of vehicle 
images, which are on the assumption of the use of helicopters an 
for vehicle manoeuvres recognition. In accordance with human 
t in high spatial and time resolution images at specific areas for 
ages certainly make observation of vehicle movement easier. As a 
r scope of utilization to derive useful traffic information on vehicle 
manoeuvres recognition with high resolution and time-serial aerial 
d stratospheric platforms. Specifically, we develop a new technique 
perception, we develop spatio-temporal clustering method, which is 
composed of geometric correction, background subtraction, shadow detection, optical flow extraction, clustering and labelling. 
Employing the two features, that are background subtraction value and optical flow, the pixels that are adjacent and have similar 
features are grouped. Consequently, vehicle clusters are formed in the spatio-temporal image. We apply the spatio-temporal 
clustering method to sequential images from a helicopter. The spatial resolution is 10cm, and the time interval of successive images 
is 1/30 seconds, and the number of frames is 600. The method yields good result, and the rates of vehicle recognition are 100%. We 
also confirm influence of spatial and time resolutions. Furthermore, we evaluate the accuracy of vehicle positions by comparing 
with the vehicle positions produced manually. 
1. INTRODUCTION 
Traffic flow surveillance for traffic control and transportation 
planning require fixed-point continuous observation of exact 
dynamic vehicle movement. Traffic flow surveillance can be 
conducted with kinds of beacons, GPS, cellular phones, video 
cameras and manual counts. Among these ways, measurement 
with video cameras is the most suitable for the continuous 
observation of the exact movement of each vehicle, which is 
important information for policy decisions on traffic problems. 
There have been some attempts to study the measurement of 
dynamic movement of vehicles with fixed video cameras (e.g. 
Kamijo et al., 2000). Because the fixed video camera’s field of 
view, however, was very narrow, the applications to traffic 
engineering were limited. 
On the other hand, images from high altitude platforms, such as 
helicopters, satisfy the requirement, that is the observation of 
exact dynamic vehicle movement in wide area. Moreover, in 
recent years, a stratospheric platform system has been projected. 
The stratospheric platform is being supposed to be kept at a 
stratospheric altitude of about 20 km. It is intended to 
contribute to telecommunication and the other different 
purposes. One of them is utilization for earth observation. 
These high altitude platforms are expected to result in high 
spatial and time resolution images at specific areas for 
continuous observation. These high resolution and continuous 
images certainly make observation of vehicle movement easier. 
As a result, the high altitude platforms have a great potential 
with wider scope of utilization to derive useful traffic 
information such as vehicle trajectories, namely locations, 
velocities, acceleration and deceleration patterns, lane changes 
and right and left turns in intersection. The information from the 
vehicle trajectories can contribute to dealing with traffic 
congestion, taking origin-destination demand proportions, 
signal control, simulation of vehicle manoeuvres, and so on. 
In this study, we explore the possibility of vehicle manoeuvres 
recognition with high resolution and time-serial aerial images, 
which are on the assumption of the use of helicopters and 
stratospheric platforms. ^ Specifically, we develop a new 
technique for vehicle manoeuvres recognition. 
2. VEHICLE MANOEUVRES RECOGNITION 
METHOD 
2.1 Ordinary Object Tracking Method 
The ordinary object tracking methods basically consist of two 
components: 
(1) extraction of objects in each image, 
(2) object tracking by making a correspondence of extracted 
objects in successive images. 
The extraction methods which employ the information of 
movement are divided into subtraction-based approach (e.g., 
Betke et al, 2000; Haritaoglu et al., 2000) and optical flow- 
based approach (e.g., Ben-Ezra et al, 2001; Medioni et al., 
2001; Smith and Brady, 1994). 
—561—