Machine Vision Based Vehicle Guidance:
Lateral and Longitudinal Control
An Image of our PATH test vehicle with the stereo camera rig mounted on top, and an image
of the interior recording equipement.
Identified vehicles traveling in the same direction as the test vehicle. On
the right, the positions of identified vehicles with respect to the
test vehicle are shown.
This project is funded by California PATH, an Advanced Technology Program that was
launched 1986 by the California Department of Transportation (Caltrans) and the
Institute of Transportation Studies of the University of California at Berkeley.
The California PATH program is committed to finding solutions to the problems of
California's current transportation systems. A major issue in this program is
to investigate the applicability of advanced highway automation using advanced
telecommunications, computing, sensors and actuators, and electronic technologies to
traffic management and vehicle control to make automobiles "smarter".
PATH is a major element of the national
Intelligent Vehicle/Highway Systems (IVHS) effort.
A key role to increase traffic capacity is the platooning concept, i.e. organize the
traffic in tightly spaced platoons. Automobiles in such a platoon will be controlled
by computers supported by advanced sensor, actuators, and communication to other
computers. The current approach is using magnetic sensors sensing
magnets in the road for lateral control and doppler radar for obstacle detection and
longitudinal control. We are investigating the possibility of supporting these
non-visual sensors using visual sensors and suggest an integrated approach.
Visual sensing becomes very important during a lane change or while merging into an
already established platoon.
Our current approach consists of using stereopsis for longitudinal measurements,
obstacle detection and to support longitudinal control. 3D measurements from the
stereo component provide the information for finding lane markers, which we detect
and track using a lane marker model in order to estimate the lateral position of the
car in the lane. We consider the following three problems in vehicle control:
- Longitudinal control i.e. maintaining a safe, constant distance
from the vehicle in front. We propose to extract information for this
using binocular stereopsis. This information can
provide an additional check for the information from the other
longitudinal sensors being studied such as Doppler radar. It also
provides a safety check for unplanned obstacles in the environment
which may not be detected by the radar.
- Lateral control i.e. road following while maintaining a constant
distance from the sides of the road. For this we propose using
line flow -- the visually measured side slip of the lane markers and
their angular rotation. This information is intended to complement and
provide a safety check for the information extracted from the magnetic
sensors currently being used. Additionally, vision provides
information about the curvature of the road which would provide useful
look-ahead for the control algorithm.
- Guiding lane changing maneuvers. Here the information provided by the
non-visual sensors is probably not adequate and vision plays a primary
role. First, we have a safety check to determine if it is safe to
execute a lane change maneuver based on the information extracted by
stereopsis. Secondly, the line flow is used as a control variable
during the lane change maneuver itself.
Further results on the use of binocular vision for highway driving. (gzip postscript 784K)
J. Weber and M. Atkin. SPIE Conference on Intelligent Systems and Controls, SPIE Vol 2902, Nov. 1996.
An integrated stereo-based approach to automatic vehicle guidance (4.8 Meg)
Q.-T. Luong, J. Weber, D. Koller, J. Malik. Proceedings of the 5th ICCV, June 1995.
Binocular Stereopsis and Lane Marker Flow for Vehicle Navigation:
Lateral and Longitudinal Control.
D. Koller, Q.-T. Luong, J. Malik. Technical report UCB/CSD-94-804, March 1994.
Dieter Koller .