Numerical processing of stereo optical images for autonomous navigation

Autonomous robots had been for a long time any engineers dream. First steps were done by creating robots able to move in well know space for restricted distances. The need of a robot able to navigate on an area with high irregularities and many unknowns obstacles lead to the organisation of the DARPA Grand Challenge. Most competitors used GPS and Laser Telemeters for terrain investigation. This paper is describing a different way to investigate the terrain topography: stereo images. This is the closest way of terrain investigation to the human sight. By gathering as much information from images the need for other sensors on the robot will be no more. This is an obvious fact since for human more than 90% of information about the surrounding environment is coming from sight. An equipment to capture and process stereo images had been build and the software was created to prove the accuracy of this method to detect and avoid obstacles and to investigate the terrain topography for uncharted areas. To extract the information needed for the navigation from the stereo optic images, complex algorithms of shape recognisance, pattern mach and cross correlation will be involved together with image calibration and sharpening. Because the processing power requirements will be huge methods to decrease the computation need will be used to increase the processing speed On site measurements will give a clear view of the performance and accuracy. For better measurements the stereo optical system will be backed up by other sensors: magnetic compass, gyroscopes, tri-axial accelerometers, pressure sensors, GPS and odometer. Some of these are redundant and are used just to confirm and improve the accuracy of the information gathered thru the stereo optical images.