Research On Positioning System Of Humanoid Soccer Robot Based On Monocular Vision

With the maturity of artificial intelligence technology and image processing technology,the research of humanoid robots has developed rapidly in recent years.Exploring the unknown environment is a relatively advanced topic in the research field of humanoid robots.This paper takes the humanoid soccer robot NAO in the Robocup standard platform league as the research object,focuses on the robot's monocular visual information perception and processing,and uses it to complete the single robot global positioning and multi-robot cooperative positioning system design,so that it can give the upper decision system.Provide effective and accurate location information.(1)Firstly,the YHS2 color space based on YUV and HSV is proposed;Sobel contour extraction is performed on each frame image in this color space,and the target shape includes arc,X-shaped cross line,L-shaped cross line,T-shape cross-line is applied,and the normalized cross-correlation(NCC)algorithm is used to calculate the correlation between the template edge gradient and the target edge gradient,and the pyramid model is used for matching optimization.To achieve the occlusion target matching,the contour of the target shape is regarded as an unordered point set extracts the shape features and uses local descriptors for local matching.The experimental results show that the average recognition rate of the algorithm reaches about 90%.(2)Secondly,Monte Carlo particle filter is used to realize the global positioning of the robot;in order to solve the depletion characteristics of the particle,the particle filter resampling step is applied to ensure the particle validity;at the same time,the unscented Kalman filter fusion is utilized.The extracted target features are locally localized,and the robot positioning is updated as an observation equation.The local positioning can reduce the update frequency of the global positioning and effectively improve the execution efficiency.(3)Finally,considering the complete symmetry of the site,the problem of robot positioning image is solved by multi-machine communication.The process is mainly divided into two steps: the realization of global site coverage and the establishment of a global ball model based on role assignment.The data type is "streamed" to form a standardization of different types of information formats;the message queue is used to read,write and store different types of stream data;and the UDP transport protocol is used for communication.The multi-robot communication positioning technology developed in this paper realizes the positioning sharing among multiple robots.