The problem of robot vision for space construction is difficult and challenging. The images that the robot vision system must deal with will be difficult to segment into meaningful pieces and difficult to interpret. Three image processing methodologies being studied at mvi include mathematical morphology, the facet model for image data, and the topographic primal sketch. The proposed project consists of 1) evaluating a number of low and mid level processing techniques employing these methodoligies and using a representative set of images that pertain to robot assembly, construction, and maintenance tasks; 2) determining how the information extracted from the images can be integrated to determine what is being viewed and in what positions and orientations, and 3) designing and implementing an integrated robot vision system using the results of 1) and 2). The phase i effort will select the test bed images, perform the experiments that evaluate toe techniques, survey the literature on integrating knowledge, and suggest a preliminary design of an integrated system. The phase ii effort will go on to complete the design and implement a working system that can handle one or more assembly or construction tasks.
