This thesis describes a vision system for recognizing man-made objects with occlusion and partial distorition. In this thesis, a new object feature primitive, local shape primitive, is introduced, and a method of object inference, selecting model objects which is probably included in input image, using the feature primitive is described. The system, described in this thesis, is a vision system using this method. Object recognition in the system, described in this thesis, is performed by two steps: the first is object inference, the second is matching. Object boundary can be represented by concurves, straight lines or circular arcs. The objects which is included in input image can be inferred with some predefined local shape primitives, characteristic subpart of object boundary represented by concurves, in object inference module. In matching module, the system identifies sensed objects by the boundary matching algorithm, matched line extension algorithm, with inferred model objects. The system enhances recognition time and improves robustness against distortion and occlusion by this method.