| Soccer robot self-localization is the basis of soccer robot games. With the research and development of computer vision and visual sensor, visual positioning as the intelligent and anthropomorphic robot location technology has attracted much attention. This thesis studies the humanoid soccer robot vision positioning system, mainly from two aspects, visual image processing and robot localization.In view of the specific scenarios of soccer robot games, where the colors and shapes of the targets are simple and known, an adaptive threshold image segmentation method was proposed, which is based on color segmentation method, so as to adapt to the illumination change. Then we used the mathematical morphology filter method to deal with the like-binarization class images, to obtain the color pieces which are closer to the real object in shape. At last, area-based seed filling method was adopted to simultaneously implement segmentation and multi-objective feature extraction. The above methods were verified through experiments for accurate and efficient segmentation of the images.Autonomous positioning of soccer robots was performed with particle filtering, with the information of visual sensing and robot movement. Relative positioning of the moving targets was realized using static geometric positioning method afterwards. Since there exists particle degradation and particle diversity contradiction in classic particle filter algorithm, and general resampling method will generate steady state inaccuracy error, an improved resampling method was proposed in this thesis, where new system information was added to enrich the particle diversity, and the particles were retained with certain noise to reduce the steady-state inaccuracy error. The feasibility and effectiveness of the proposed particle filter algorithm was illustrated by simulation and experiments, and the robot positioning and relative positioning of the moving targets were realized on the humanoid soccer robot platform. |
PDF Full Text Request