Research Of Mid-Sized Soccer Robots' Vision Recognition And Vision Localization

Robot soccer is an exciting research area aiming at driving forward robotics and multi-agent research through friendly competition. Designing a soccer robot is involved in machinery, electronic, automatic, pattern recog- nition, machine vision and artificial intelligence. The system of soccer robots must be provided with highly real-time and accurate. Vision is one of the important approaches of acquiring environment information, its performance is critical to improve the function of the whole system. At first, we design an omni-directional mirror in accordance with the requirements of middle-size soccer robot, then environment information be acquired with a field of view up to 360o through this mirror and The algorithms of motion objects detection, localization and tracking in image have been studied in this thesis. Since the competition field is tagged by specified colors. Some algori- thms of color image segmentation are proposed in thesis. These algorithms use the YUV color space and translate color distribution into binary image based on the YUV color space. So it reduces effectively the quantity of image infor- mation and image processing time. Using the methods of image recognition to collect the video image and to process it on time, it has been achieved that motion objects be detected and located under the complex background. Based on the theory analysis and experiment, research about the real-time algorithm of motion objects recognition and localization in video image and also the hardware device has been completed. At last, we program the software about image object recognition and localization using Visual C++ in PC platform. Then we process the image which be acquired through the vision system using the software. The experi- ment of controlling the movement of soccer robots using the result of image processing indicates the validity of our algorithms.