Research On Visual Recognition And Tracking Of Soccer Robot Based On RoboCup

With the development of artificial intelligence and robotics explosive development,computer vision technology has become a research problem in all areas of daily life and industrial production,the urgent need to break through,is of great significance to promote the RoboCup both R & D and comprehensive education international research project on robot vision technology.In this paper,the visual recognition and tracking of RoboCup small soccer robot are improved and implemented in two aspects.This paper RoboCup soccer robot for illumination changing,target motion speed and obtain the actual image distortion unfavorable factors of competition,improve the recognition of the problem of poor performance of the first,the local invariant feature BRISK algorithm in-depth study,proposes a fast and robust local feature matching algorithm based on scale invariant,the differential adaptive corner detection method for accelerated segmentation(DA-AGAST),combined with the key point fast and robust scale invariant description method(SUBRISK),the targets in complex scenes,the target feature generation with strong robustness,the realization of RoboCup robot on the sphere target fast and precise recognition.Then,according to the RoboCup soccer robot soccer system has significant nonlinear characteristics,resulting in visual tracking accuracy and efficiency of the poor,and the presence of occlusion and distortion factors in soccer robot vision tracking performance robustness problem,propose a method based on the glowworm swarm optimization particle filter algorithm hybrid strategy using the optimization characteristics,optimization algorithm,the algorithm precision and robustness are improved,and the method is applied to the RoboCup sphere target tracking.In this paper,the visual recognition and tracking performance of RoboCup soccer robot is enhanced by the two improvements.The specific contents of this study are as follows:Firstly,aiming at the poor robustness of binary locality invariant feature algorithm for target recognition problem,a local feature matching algorithm based on fast robustness scale invariance is proposed.The algorithm uses a differential adaptive corner detection method for accelerated segmentation(DA-AGAST),the rapid generation of detector has strong affine invariance,and introduce the key points of fast and robust scale invariant description method(SUBRISK),reduce the rotation scaling effect on feature matching,and through adjusting the sampling distribution and storage methods,improved algorithm the efficiency,and the method is applied to target recognition.Secondly,aiming at the poor accuracy of particle filter for target tracking problem,a firefly optimization particle filter algorithm(MSFA-PF)based on hybrid strategy is proposed.The algorithm introduces chaotic perturbation search strategy to make the particle get full search ability at the global optimal position and move efficiently to the high likelihood region.Then the dynamic visual search strategy is used to enhance the search efficiency of the particle in the high posterior probability density region.Finally,by improving the fluorescence brightness update mechanism,we enrich the particle optimization set to improve the overall quality of particles in the iteration process,and enhance the performance of particle filter to a large extent,and apply the above methods to target recognition.