Context Transformations of Range Images Using Directional Data Back Fill Algorithm

Abstract

This paper discusses the subject of filtering range images constructed using vision systems. Construction of a range image using the laser triangulation method LTM is burdened by noise resulting from the specificity of the adopted method. This method uses structural illumination, usually in the form of a laser line from the object illumination. The image of the object is captured by a camera, whereupon the laser line image is transformed into height profiles, which are then used to construct a range image of the object. Yet, such images fail to describe the entire surface of the object, due to occlusion. Occlusion of image fragments occurs in two forms. The first one stems from the impossibility to fully illuminate all surfaces of the object, due to occlusion of the laser lines. The second form occurs when the surface is illuminated properly, but the image of this surface cannot be captured by the camera. In both cases, there is an occlusion which usually results from the shape of the object. This paper describes the method of backfilling the data in undefined areas of range images caused to the phenomenon of occlusion. It suggests a method of defining the direction of data backfilling, identification of undefined areas, as well as methods of calculating and backfilling height differences at undefined spots. The purpose of these transformations is to prepare the image for measurement and control tasks. Select examples are presented with the results of image transformation using the developed data backfilling method.