The identification of objects is a difficult task because the objects of the real-world are highly variable in aspect, size, color, position in space, etc. The system of identification of object must thus have a very great adaptability. In this article we present a system of identification of the type (model) of vehicles per vision. Several geometrical parameters (distance, surface, ratio ... ) of decision, on bases of images taken in real conditions, were tested and analyzed. According to these parameters, the rate of identification can reach 95% on a basis of images made up of 9 classes of the type of vehicles. The fusion of the three classifiers using the rate of identification for each parameter allows showing the effectiveness of our process for the identification of the type of vehicle.