Seite wählen

A.3.6.4b Recognition of 3-D vehicle shape and trajectory driven

[Schick 1992] has parameterized about a dozen shapes of different car types and of vans in a common scheme in an attempt to classify vehicles actually seen from any aspect condition. The approach proposed for estimating both relative vehicle state and shape parameters used

  • edge features from a graphic display of the scene simulated by CGI (for typical car shapes see left).
  • A stationary camera fixated the simulated vehicle while it drove on a course consisting of two semicircles and two straight sections between them (see dotted oval shape in figure).
  • Using the model of Ackermann steering, beside the trajectory driven also the lateral control input (steer angle, or curvature) was to be estimated.

In this academic investigation it could be shown that the approach works, in principle. In realistic environments, reflections on real cars with curved surfaces (and a large percentage of them with very good reflection properties) a much richer set of visual features will certainly be required.

The graph to the left shows that curvature C of the trajectory could be well recovered from the model underlying the observation process; without this model of Ackermann steering and observation over time with the differential equation constraints for the dynamic process, this would be much harder.


Schick J, Dickmanns ED (1991). Simultaneous Estimation of 3-D Shape and motion of Objects by Computer Vision. IEEE Workshop on Visual Motion, Princeton. pdf

Schick J (1992). Gleichzeitige Erkennung von Form und Bewegung durch Rechnersehen, Dissertation, UniBwM / LRT. Kurzfassg

Dickmanns ED (2007). Dynamic Vision for Perception and Control of Motion. Springer-Verlag, London (Section 14.6.1)