As already indicated, the OF vectors which belong to an accepted OIC are projected back into a plane parallel, but above the road plane in the scene. Figure 7 presents a metric sketch of the gas station premise shown in Figure 34. The resulting cluster of vectors is enclosed by a rectangle, see Figure 8. The average scene velocity estimate for the OC obtained from this cluster is computed in order
Figure: Metric sketch of the gas station premise shown in Figure 34.
Figure 8: OC which has been obtained by back projecting and clustering the OF vectors. The OC and the OF vectors from which it has been generated are superimposed to a digitized map of the petrol station premise. In the lower left quadrant, one can recognize the right part of the elongated island on which the petrol pumps are placed.
The scene position and orientation of the OC obtained in this manner is combined with an - so far still interactively selected - polyhedral object model (Figure 11) in order to provide an initial instantiation of a vehicle to be tracked in the scene, see Figure 12.
Figure 9: We use a parameterized 3-D generic model to represent the various types of vehicles moving in traffic scenes. The model comprises 12 length parameters.
|Bottom Length||bl||Length of the bottom area of the car.|
|Bottom Width||bw||Width of the bottom area of the car.|
|Bottom Heigth||bh||Height above the ground of the car.|
|Roof Length||rl||Length of the car roof.|
|Roof Width||rb||Width of the car roof.|
|Roof Height||rh||Heigth of the car roof (considering the ground).|
|Roof Edge||re||Distance between the front of the roof and the front of the car.|
|Front Length||fl||Length of the engine bonnet.|
|Front Heigth||fh||Heigth of the front of the car.|
|Stern Heigth||sh||Heigth of the stern of the car.|
|Body Heigth||boh||Height of the car body.|
Figure 10: Example of five different vehicle models derived from the same generic model.
Figure 11: A polyhedral object model
Figure: The scene position and orientation of the OC is combined with a polyhedral object model in order to provide an initial instantiation of a vehicle.
Standard Computer Graphics techniques are used in order to project the instantiated vehicle model back into the image plane. Based on an interactively obtained estimate of the direction of incoming light, the shadow cast by the instantiated model onto the road plane can be computed ([Koller et al. 93]) as shown in Figure 17.
Figure 13: Example of a video image frame with distinctive shadows.
Figure 14: The intersection of a light ray through 3-D model vertex and the street plane leads to the vertex projected on the street plane.
Figure 15: Adaption of a car model without considering the shadow.
Figure 16: Adaption of a car model considering also the edges of the shadow.
Figure 17: Projection of the instantiated vehicle model into the image plane, including the projection of the shadow cast by the vehicle model onto the road plane. The (two-parameter) unit vector characterizing the incoming light direction in the Scene Domain (SD) has been determined interactively so far.