Video based indoor exploration with autonomous and mobile robots

Techniques are proposed to support the video based development of systems for indoor exploration with mobile robots. The technique of redundant programming is often used to improve the reliability of operating systems, but the use of this technique is not common for CV (computer vision) applications. Also a new technique to create CAD (computer aided design) models from image data is described. These techniques were used for the development of an RV (robot vision) program. The observed recognition power exceeds the abilities of sophisticated but conventional programs clearly. This is documented with sample images, which show a table that has been taken from different distances. The quality of the images is very bad due to the fact that a camera was taken which has a very low resolution. Additionally the detection of the table was hampered, because the illumination in the images varied considerably. Sometimes the table was placed very near by a window with strong exposure to sunlight. Over-exposure of the table complicated the reconstruction because of this problem. Sometimes other objects irritated the detection. The program handled all these difficulties impressionably although it used no calibration techniques. No other robot-vision program is documented in the literature that gained the reported recognition rate.