VEHICLE DETECTION IN TRAFFIC IMAGES
Luigi Di Stefano DEIS, University of Bologna
Viale Risorgimento 2, 40136 Bologna, Italy
and
Enrico Viarani DEIS, University of Bologna
Viale Risorgimento 2, 40136 Bologna, Italy e-mail
Topic Area: Artificial Intelligence and Decision Support Systems
Keywords: Applications of Pattern Recognition, Computer Vision, Traffic
Monitoring
Extended Abstract
This paper describes the research activity currently going on at DEIS,
University of Bologna, on the deployment of computer vision techniques
for extraction of traffic information from images. The perspective goal
of this activity is the development of a low-cost, vision-based traffic
monitoring system capable of measuring fundamental traffic parameters,
e.g. queue length at intersections, as well as recovering higher level
information such as for example assessment of the traffic situation and
congestion detection.
In this framework, one basic problem consists in the development of
fast and robust algorithms for vehicle detection and tracking.
According
with related work, we have chosen to rely on the extraction of
motion
from image sequences so as to segment out vehicles with respect to
still
objects (e.g. road, buildings..). In particular, our approach is based
on the Block Matching Algorithm (abbreviated as BMA), which used for
motion
estimation in the MPEG image compression standard. BMA has been
employed
for vehicle detection in an application aimed at evaluating turning
rates
at urban crossroad. Unlike techniques based on spatio-temporal
derivatives
or background subtraction, BMA provides not only detection of moving
points
but also motion estimation (i.e. a motion vector is associated with
image
points). This improves the discrimination power in case of partially
occluding
vehicles and crowded traffic scenes. In addition, specialised hardware
for BMA-based real-time motion estimation is available nowadays.
Get the paper postscript file
