An approach for the model-based tracking of vehicles in road traffic scenes has been examined in detail in an extensive analysis. As a consequence, more effort was spend on the improvement of the initialisation phase. By applying knowledge, which is utilized in the tracking phase anyway, the initial estimation of the vehicle’s position could be determined more robust. For this, we exploit the shape of an vehicle as well as its shadow, which is cast on the street plane. A quantitative analysis verifies the achieved improvements.