In many areas of application, such as medicine, robot technology, photogrammetry, etc., the acquisition of an abstract description of three dimensional objects is an important task. Our approach is directed to an integrated system called SMART (Segmentation Matching And Reconstruction) that is based on general purpose equipment (general purpose workstation, photographic camera or CCD camera). It is designed as a self-calibrating system, i.e., the camera positions, as well as their relative orientations, are derived automatically during the evaluation of the image series. Hence the photos need not be taken by a specially trained person. The whole procedure within the SMART system can be reflected in a vision pipeline (see section). After the image acquisition we perform a rough segmentation of the images (1) to find characteristic geometric primitives of the objects and to reject non characteristic ones whose occurrence is unavoidable during the acquisition process and (2) to calculate the camera positions approximately. Based on this information we measure the exact positions of the characteristic details and their correspondence. This data allows the usage of self-calibrating reconstruction algorithm. Now, the obtained partial reconstructions are connected to one complete reconstruction.