7A-5-1 
Traffic Sign Detection from Image Sequences 
W. B. Tong J. Y. Hervé P. Cohen 
Groupe de Recherche en Perception et Robotique 
Ecole Polytechnique de Montréal 
Montréal (Québec), Canada, H3C 3A7 
Abstract 
This paper describes automatic traffic sign detection 
based on colorT shape and text information of traf 
fic signs for mobile mapping. Vision technologies are 
applied to extract georeferenced data to build a geo 
graphic information system. This gives georeferenced 
traffic signs data for use in a highway pavement man 
agement system. 1 
1 Introduction 
Geographic Information Systems (GIS) have many ap 
plications. An example is map-making [1] for highway 
and transportation infrastructure with the identifica 
tion and location of road signs based on stereo vision 
and on Global Position System (GPS) technologies. 
To collect dataT vehicles like the one illustrated in 
Figure 1 are used. This one basically consists of a 
GPS and a stereo camera system mounted on a van. 
While the van is traveling at a normal traffic speedrthe 
stereo camera system records stereo video sequences 
for a roadway. Each video frame is tagged with the 
GPS signal and geodetic coordinates: latitude, longi 
tude and ellipsoidal height. The video sequences are 
post-processed to detectridentifyTand locate road fea- 
turesTfor instancerroad edges and traffic signs. 
Existing mobile mapping systems typically run 
with an operator interactively clicking on objects of 
interests in video sequences to show the system where 
the object is and what it is. To speed up data col 
lection and reduce operator workloadTit is natural to 
develop an automated system for GIS objects extrac 
tion from video sequences. 
1 This research work is supported by the National Research 
Council of Canada and in collaboration with GIE Technologies 
Inc. 
Figure 1: A van with a stereo camera system and GPST 
courtesy: the Lambda Company 
2 Related works 
At the Image Recognition LaboratoryT Universität 
KoblenzrGermanyra real time traffic sign recognition 
system has been developed in cooperation with Daim 
ler Benz as part of the European PROMETHEUS 
project (Program for a European Traffic with High 
est Efficiency and Unprecedented Safety) [4]. For re 
altime analysisTthis system employs special hardware 
to process traffic images. In the same projectTat the 
Ulm Research center of Daimler-Benz AGTa hybrid 
approach has been developed for traffic sign recogni 
tion. It involves color segmentation with a pixel clas 
sifier and a pictogram classifier [5]. 
At the Department of Electrical EngineeringlTexas 
A. & M. UniversityT USATa traffic sign recognition 
system [3] has been developed to alert drivers of ap 
proaching warning signs: ‘stop’FyieldTand ‘don’t en- 
terTetc. 
A robust road detection and recognition systemT 
developed at the University of GenovaTworks in three 
stages. The first stage reduces the search of a road 
sign to a subregion of the image using a priori knowl 
edge of the scene or color clues. The second stage 
geometrically analyzes edges extracted from images 
to identify regular sign shapes. The last stage uses 
a cross-correlation technique to recognize traffic signs 
[2].