Research On Target Recognition And Tracking Of Soccer Robot Based On Omnidirectional Vision

Visual information processing technology is the most basic and most critical technology for football robots to complete various tasks.The research background of this paper is based on RoboCup medium-sized soccer robots.The research on medium-sized soccer robots includes visual design,motion control,hardware design,path planning,attack and defense strategies.Among them,the omnidirectional vision system is in a critical position,and the performance of the system directly plays an important role in whether the soccer robot task is successfully completed.The research on this system can promote the development of related technologies in the field of artificial intelligence,and at the same time make the application of machine vision more diverse.In the application scenario of the medium-sized soccer robot competition,the target soccer is the most important recognition task of the soccer robot.When the target is recognized,external disturbances such as light intensity,image distortion,etc.are generated.Therefore,the anti-interference and recognition accuracy are aspects that the target recognition and tracking algorithm mainly needs to improve and improve.This paper focuses on the following aspects of research and improvement of soccer robot target recognition and tracking:(1)Image preprocessingThe visual system of medium-sized soccer robots generally adopts an omnidirectional vision system.The color space of the system needs to be selected.This paper combines the components of HSI color space and YUV color space to design a more targeted The color space is more conducive to the recognition of the target.In the process of image preprocessing,the homomorphic filtering algorithm is improved,and the algorithm is combined with Mallat wavelet transform to make up for the shortcomings of traditional homomorphic filtering for the enhancement of local features.The improved method can improve the signal-to-noise ratio of the image,and can completely preserve the local information in the original image.(2)Recognition of TargetIn the process of identifying the target object,it is necessary to scan the image in an area.The general method is to scan pixel by pixel,which is relatively inefficient.Moreover,the target recognition of soccer robots generally adopts the threshold method,but the method is sensitive to illumination.So in order to overcome these shortcomings,this paper proposes an SVM algorithm based on biSCAN scan line.The experimental results show that the proposed algorithm is robust and has strong adaptability to illumination changes,which can meet the needs of the game.(3)Tracking of targetIn this paper,mean shift is used as the theoretical basis of target tracking,and the target feature model is improved for the characteristics of omnidirectional vision,which makes the improved algorithm more suitable for target tracking of soccer robots.The algorithm has strong anti-interference to the change of target shape and angle,and is suitable for target tracking in omnidirectional vision.The experiment proves that the algorithm has a good tracking effect on the target and basically meets the requirements of target tracking in the game.Throughout the whole article,this paper improves some shortcomings of traditional algorithms,and proposes a homomorphic filtering image enhancement algorithm based on Mallat wavelet transform,a SVM recognition algorithm based on biSCAN and a target feature model based on omnidirectional visual features.Experimental results show that These improvements effectively improve the accuracy and stability of target recognition and tracking.