Reconstructing 3-D objects from image data is one of the most important problems, as the first step to make robots which can recognize environment and work by themselves for handling materials, grasping and manipulating parts and so on. As the basis on reconstructing 3-D objects, a new method for deciding edges in images is presented. This method consists of the following three subprocedures : (1) Differences of brightness of two subsequent pixels are calculated and binary images are produced, (2) Noise included in those image data is roughly eliminated by utilizing a property of an edge in such images, (3) Noise not recognized in the previous subprocedure is eliminated by the relation between 3-D objects in 3-D space and images, by using multi-image data obtained by the camera moving on a rail. An experimental camera system has been constructed, in order to confirm the theoretical consideration, and various examples have been tested. The results show that most of noise can be recognized and eliminated, and positions of edges in 3-D space can be reconstructed.
