Design And Realization Of Hawkeye System In Curling Competition

With the continuous development of science and technology,the field of artificial intelligence has also ushered in a new period of development,and has produced more and more cross-applications with the sports industry.At present,the vision system with artificial intelligence technology as the core is applied more and more widely in the sports industry.This article takes curling competition as the research background,and builds a Hawkeye system that can be used in curling competitions based on artificial intelligence related technologies.The system can complete tasks such as curling venue calibration,curli ng target detection,and curling track preservation and playback.The relevant implementation is as follows:First of all,the field of view calibration for the curling scene is realized.For the multi-camera system on the curling field,the correction of the camera distortion parameters has been completed.Aiming at the problem of large field of view calibration,a site calibration scheme based on April Tag was proposed,which theoretically verified the feasibility of the scheme and carried out a specific s cheme design.It is verified through experiments that the calibration scheme based on April Tag has a smaller error than the calibration scheme based on curling scene information extraction features,and can reach a positioning accuracy of 0.05 m.The calibration results of this venue can be used for subsequent global positioning of curling.Secondly,the target detection of the curling ball in the track is realized.In this paper,the curling data set is constructed by collecting curling image data on the s pot,and the data set is enhanced according to the environmental information of the curling scene.Based on the curling data set,the training and testing of the target detection network is completed.The mean average precision on the test set can reach 99.6.At the same time,this article also researches and analyzes the network performance based on Tensor RT acceleration and network sparse acceleration.The experimental results show that the performance of the network model base d on Tensor RT acceleration is better.While the inference speed reaches 58.9 frames per second,it still maintains the mean average precision of 95.7%,which can achieve a balance between accuracy and speed and meet the performance indicators of the system.Finally,complete the overall design and analysis of the Hawkeye system.Through the analysis of system requirements,the overall architecture is designed.Starting from the actual scene of the curling hall,the camera parameters that meet the actual site are obtained through calculations,and the selection of the camera is completed,and on this basis,the hardware part of the Hawkeye system is built.The software is partly based on the QT framework and developed in a modular manner,which can better interact with users.Through actual tests,each functional module runs successfully,the system runs at speed of 50 frames per second,and the curling positioning accuracy is within 0.1m,which can provide effective information for curling players’ training.