obstacle detection and avoidance
obstacle and stopping in front of it 1988
Avoiding obstacles is
very essential for safe driving; in monocular vision, the easiest
clue to distance of an object on the ground in the driving path is
to assume a planar surface and to determine range by the lowest
coordinate of a set of several (horizontal) edge features
belonging to the same object.
This property may be
checked by a homogeneously bright or dark region between the edge
The distance to the
object is then given by the row index of the lowest feature,
assuming that gravity pulls the object to the ground.
Once a set of outer
edges has been found, commanding the vertical search in the next
image at the center between the horizontal features in the
present image leads to tracking of the object vertically with
relatively little effort;
the horizontal search at the center between the vertical features
leads to horizontal tracking.
coordinates found approximately mark the center of the obstacle
area; range is derived from the row coordinate of the lowest
feature in the vertical.
Final demo of project
“Autonomous Mobile Systems” in Rastatt, 1988:
Stopping in front of a
trash-can detected by vision at a speed of ~ 40 km/h.
Stat.Obstacle Rastatt 1988
Dickmanns ED, Christians T (1989).
Relative 3-D-state Estimation for Autonomous Visual Guidance of
Road Vehicles. In T. Kanade et al (eds): 'Intelligent Autonomous
Systems 2', Amsterdam, Vol. 2, pp 683-693; also appeared in:
Robotics and Autonomous Systems 7 (1991), Elsevier Science Publ.,
Dickmanns ED (2007). Dynamic
Vision for Perception and Control of Motion. Springer-Verlag,
controlled distance keeping to vehicle
2a) Single vehicle
coming to a stop: In the early 1990’s, radar and vision
had a competition as to which approach is the better one for
improving safety in obstacle avoidance in dense traffic or even a
traffic jam. At that time, the computing power available at
acceptable costs did not allow considering vision as economically
viable; also, the state of micro-processor technology and image
evaluation software did not allow reliable detection of any kind
of vehicles under all conditions.
Nonetheless it could
be proven that even monocular vision was able to solve the task
under normal conditions. The figure to the left shows
a comparison in range
(bottom) estimated with
a mono-beam laser
range finder, hand-pointed to the vehicle of interest (thin
the 4-D monocular
dynamic vision system (heavy
The agreement is quite
good. First results of visual ‘Stop & Go’ have
been demonstrated at the demo Prometheus 1, Torino, 1991.
Thomanek F, Dickmanns D (1992).
Obstacle Detection, Tracking and State Estimation for Autonomous
Road Vehicle Guidance. IEEE/RSJ Int. Conf. on Intelligent Robots
and Systems, IROS) Vol. II, Raleigh, pp. 1399-1406
Schmid M, Thomanek F (1993). Real
Time Detection and Recognition of Vehicles for an Autonomous
Guidance and Control System. Pattern Recognition and Image
Analysis, Vol. 3, No. 3, pp 377-380
Brüdigam C (1994).
Fahrmanöver sehender autonomer Fahrzeuge in autobahnähnlicher
Umgebung. Dissertation, UniBwM /