This knowledge about the camera is exploited in order to project OF vectors from an accepted OIC cluster into a plane parallel to, but above the road plane at the average height of a vehicle. This plane is part of the Scene Domain (SD) (see, e. g., [Kanade 78]) as opposed to the PD within which the OIC has been hypothesized. The OF vectors in this SD-plane are re-clustered and enclosed by a rectangle which is attributed by the average velocity parallel to the road plane as obtained from the OF vectors which have been accepted into the OIC.
Based on the projection of this SD-rectangle onto the road plane, a hypothesis is formulated that a road vehicle of a certain type (currently still determined interactively in the Xtrack system) happens to be placed at the given coordinates in the scene at the time point corresponding to the current (half)frame. It is assumed that the vehicle drives forward, i. e. that the front part of the vehicle is aligned with the side delimiting the SD-rectangle in the direction of motion as indicated by the velocity vector attribute. The resulting hypothesis is denoted as Object Candidate (OC). The concept of OC relates to the SD, not to the PD as the concept of Object Image Candidate (OIC).
The representation of the shape of an object type could be parameterized as explained, for example, in [Koller et al. 93].
Since at least the position, orientation, and velocity of a vehicle are parameters to be estimated during the initialisation phase, the system-internal representation of an object type will be referred to as a generic object description (GOD). In the case where some or all shape parameters have been specified, for example interactively as in Xtrack so far, we shall talk about a (partially) specialized GOD. An example is given in the top component-figure of Figure 4, based on a polyhedral GOD.