Object positioning and 3D modeling is not only a promising research direction in robot welding, but also an emphasis research field of machine vision. For objects lacking of textures on its surface, such as welding part, problem of 3D modeling of this kind of object can not be settled by traditional binocular stereo vision. In this paper, an approach to the problem is presented. A multibaseline stereo vision system with four verging configured cameras is proposed. Experiments show that we can get dense data points by the method, the results of surface reconstruction of objects with no matching texture on its surface is good. Experiment results confirmed that the method is valid in solving the problem of 3D surface reconstruction of objects lacking of matching texture.