Research On Soccer Robot Target Recognition Based On Omni Directional Vision

With the in-depth study and mining of intelligent algorithms,robotics has developed rapidly,and the robot used as an intelligent algorithm is becoming more and more intelligent and humanized.How to further enhance the learning ability of robots and make them more intelligent is the direction of the industry.Soccer robot is a good platform to test the advantages and disadvantages of intelligent algorithms.As one of the most influential robot soccer competitions in the world,RoboCup is a typical multi-agent collaboration system,such as robotics,computer graphics,communication and so on.The middle size soccer robot is a complex multi-agent system including motion control,path planning,visual design and hardware design.The visual system of soccer robot occupies an important position.The real-time and accurate recognition of the vision system will directly affect the decision planning,the formation selection of the decision system and even the result of the game.In the visual system,the task of target recognition is mainly borne by the omnidirectional vision system.Therefore,it is the focus of this paper to improve the real-time and accuracy of the target recognition of the omnidirectional vision system of the soccer robot.The medium group soccer robot mainly uses the omnidirectional vision system to complete the image acquisition and target recognition.On the basis of reading a large number of domestic and foreign literature,and on the analysis and comparison of the algorithms proposed by domestic and foreign scholars on target recognition,the advantages and disadvantages are pointed out.This paper mainly deals with the real time recognition of ball in the field.Sex and accuracy were studied.The main work is as follows:(1)Aiming at the problem of ball recognition in soccer robot competition,a BP neural network algorithm is proposed for target recognition.In the middle size soccer robot competition,there are specific colors in the competition environment and field targets,for example,the green space and the orange color.Therefore,this paper mainly starts with color recognition.First,three common color models,RGB,YUV and HSV,are analyzed,and the RGB color model with low time complexity is selected.Then the traditional BP neural network is improved by introducing momentum factor and adaptive learning rate to identify the target.(2)In view of the problem that the recognition accuracy of the soccer robot target recognition algorithm is not high and the algorithm runs too long,a target recognition algorithm is proposed,which combines the color feature and the random Hough transform.The algorithm performs color filtering in the R channel in RGB space,and combines the Otsu adaptive threshold method to segment the image threshold,and then uses the improved random Hough transform to recognize the target.