3D model reconstruction plays a very important role in computer vision as well as in different engineering applications. The determination of the 3D model from multiple images is of key importance. One of the most important difficulties in autonomous 3D reconstruction is the (automatic) selection of the "significant" points which carry information about the shape of the 3D bodies i.e. are characteristic from the model point of view. Another problem to be solved is the point correspondence matching in different images. In this paper a 3D reconstruction technique is introduced, which is capable to determine the 3D model of a scene without any external (human) intervention. The method is based on recent results of image processing, epipolar geometry, and intelligent and soft techniques. Possible applications of the presented algorithm in vehicle system dynamics are also presented. The results can be applied advantageously at other engineering fields, like car-crash analysis, robot guiding, object recognition, supervision of 3D scenes, etc., as well.