Develop a system based on omni-directional vision for navigation of a mobile robot. The system will perform in an a priori known limited environment of a robotic contest with several kinds of artificial visual landmarks. The task is complicated by the presence of additional objects (such as opposing robots) in the operation field and frequent interactions influencing the desired robot motion. The system must deal both with frequent view occlusions and random displacement. 2)
The thesis addresses the vast problem of mobile robot localization in the dynamic environment of a robotic contest. A method based on particle filters using only the image of a catadioptric visual sensor as its input is developed. The advantage of this approach is increased robustness to external influences such as robot collisions and easy portability thanks to independence on other systems of the mobile robot. The proposed method employs a composition of fast color thresholding, look-up coordinate transform, vision-based motion prediction and Monte Carlo Localization to gain robust and reliable pose tracking using a color map of a delimited environment. Since the system uses visual data both to determine the relative motion and to verify the current location, it can cope well with unexpected events such as wheel slippage or collision.