The Design Of Multi-agent Visual Positioning Sysytem And Positioning Algorithm

Multi-agent system is one frontier topics of artificial intelligence and cooperativecontrol research, which has the characteristics of intelligence, collaboration and a goodability of fault tolerance. Due to the complexity of modern control systems, individualcontrol can’t meet the requirements, cooperative control technique of multi-agent systemshas been more and more applied in the actual industrial systems. Positioning is one of theimportant functions for agents completing a task cooperatively. Thus it is significant tostudy positioning system and positioning algorithm for multi-agent systems. This paperfulfills the following work:Firstly, a multi-agent positioning experiment system is built to provide a practicalsimulation platform for multi-agent system control algorithms. This system is composedof visual positioning subsystem, motion subsystem, decision subsystem and datacommunication subsystem. We design a multi-agent positioning software for visualpositioning subsystem to complete the function of positioning; the motion subsystemconsists of several color-marked SRV-1mobile robots; the decision-making subsystemconsists of DSP processors of SRV-1mobile robots, which store the multi-agent controlalgorithms; and the data communication subsystem is established by indoor wireless localarea network.Secondly, the HSV color model positioning algorithm and positioning software aredesigned. Positioning algorithm transforms RGB color model to HSV color model toprocess color information, and this algorithm reduces the interference of light of outsideenvironment in positioning process. VC6.0is used to developing software interface, theproposed positioning algorithm is applied in software, UDP protocol is used to completecoordinate data information interaction.Thirdly, the improved Camshift algorithm is proposed. The algorithm contains thesearch window weights right value function, the kalman filter and contour matchingalgorithm, which reduces the positioning error of the target shading and obstacles withsimilar color in actual positioning process. The experimental results show that this methodis effective.Finally, the multi-agent formation experiment is designed to finish the adjustment ofthe positioning system. First, the single agent to fixed-point experiment is designed: thefixed-point control algorithm is loaded to each agent’s underlying decision module, and the agent received the real-time positioning information transferred by positioningsoftware to complete the experiment. Then, the tracking and intelligent containmentexperiments are designed on the basis of fixed-point experiment: with the change of targetagent, the tracking agents choose the corresponding control strategies to complete thetasks of real-time target tracking and best position containment. The above experimentsprove the stability of the positioning system and the accuracy and instantaneity of thepositioning algorithm.