Principal Investigator

Dr.LE Thanh-Sach
Ho Chi Minh City University of Technology (HCMUT), Vietnam

ASEAN Co-Investigator

National University of Laos (NUOL)
Lao PDR

Japanese Co-Investigator

Prof. Dr. Kazuhiko HAMAMOTO
Tokai University

Awarded year

2014

Program

Collaborative Research Program for Common Regional Issues (CRC)

Field

Computer and Information Engineering

Abstract

“traffic parameters are important and useful. There are several existing
approaches to the collection problem. According to [1] and [2], the existing approaches can be classified into two
groups: “in-situ“ and “FCD“. “in-situ” approach needs to install collectors statically on road’s surface, under
road, or on road-side. There are two sub-groups in “insitu” – “instructive” and “non-instructive”. With instructive
approach, collectors are installed under road, i.e., roads are needed to reform. Meanwhile, non-instructive
approach only places collectors on road or on road-side. FCD (Floating car detector) collects parameters using
mobile devices installed on vehicles or wore by drivers. GPS’s data or mobile signals are used to obtain the
location of vehicles/human and to determine the density or vehicle count.
As compared in [1], data obtained by “FCD” is highly accurate; but, it is difficult for FCD to obtain dense
data. The reason of this fact is that it is not the case that all vehicles using GPS and not the case that all drivers
bringing mobile phones. Even GPS is available, it’s data is not always shared because of privacy. Especially, in
the case of Vietnam and Laos, almost vehicles are motorbike and they are not installed GPS inside. So, it is
unreality to collect traffic parameters by using “FCD” in Vietnam and Laos.
As summarized in [3] and [4], vision-based collectors have demonstrated its successfulness, at least for
large-size vehicles. Especially, recent researches on using Learning-based approach [5] and on using local visual
feature [6] showed that it is possible to predict or to estimate the number of objects (human, vehicles) by learning
the mapping between visual features to the object count or by training classifiers with local features. Therefore,
we approach the collection problem by using traffic cameras, which are able to be installed on road-side, i.e.,
flowing non-instructive approach. Moreover, there are already around 200 cameras in Ho Chi Minh City
(Vietnam). Under the relationship between HCMUT and local government, it is possible for us to request for
experiment data. It is also easy for us to construct a mobile video collectors (i.e., video acquisition) on road-side,
and we will use them to collect traffic videos in Laos or any other site.
Based on the above explanation, the purposes of this proposal are clearly defined as follows:
(1) To propose and to implement an open framework, which is collection of algorithms, for estimating the
following parameters from traffic cameras: (a) vehicle count, (b) road’s density, and (c) traffic flows.
The proposed framework is able to work with traffic conditions in Vietnam and Laos, i.e., workable
with small-size vehicles like motorbike and bicycle and workable with highly occlusion situation
among vehicles. We also aim to perform the estimation in real-time, which is for using with broadcast
systems.
(2) To research on the way to produce traffic parameters by cooperating many cameras watching on
different view-points for a large scene. Basically, one camera can observe only a small part of a large
scene. Therefore, in the case of a large scene, for example, a large roundabout, it is necessary to
combine the output of many cameras to perform the estimation.
(3) To develop a prototype for collecting traffic parameters, which uses the afore-mentioned framework.
The prototype will provide a graphic-user-interface to users for managing camera network, for
operating all kind of tasks in traffic parameter collection, and for evaluating different solutions to
parameters collection. We will create a benchmark for two approaches, “direct” and “indirect”. Based
on the comparison results, we will select the right method for collecting parameters.”

Project at glance

-