Font Size: a A A

Research On The Recognition System Of Table Tennis And Batter Based On Computer Vision

Posted on:2019-12-20Degree:MasterType:Thesis
Country:ChinaCandidate:L L DaiFull Text:PDF
GTID:2428330548978831Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
The key technologies of table tennis robot vision system such as visual based measurement and location,table tennis force analysis and trajectory prediction,batting strategy and planning,have widespread application prospects.This thesis proposes a new kind of vision system of table tennis robot,which can not only recognize and locate moving table tennis for hitting back but also identify the player's identities from dynamic gestures.The main contributions are:(1)The real time vision based recognition and tracking methods of table tennis.Aiming at the bad robustness of the current recognition method based on color analysis,a rapid algorithm for table tennis recognition based on the fusion model of RGB and HSV is proposed.We also propose a region of interest(ROI)algorithm to improve the real-time performance by processing in the local image parts rather than in the global image one.Experiments show that by using this method to track and locate the table tennis ball,the positioning error is less than 20 mm and the algorithm processing time is less than 10 ms.These performance metrics meet the requirements of practical hitting of table tennis robot.(2)A robust and accurate batting prediction strategy.By analyzing the flight and collision models of table tennis ball,the initial velocity feedback regulation algorithm is designed for trajectory prediction.And we propose a calculation method for hitting parameters such as hitting point,racket attitude and swing speed.In real experiments,the industrial six-axis robotic arm can achieve the ball hitting action by using this kind of batting prediction strategy.(3)The exploration of dynamic gesture based batter identification.Due to the importance of gesture recognition in the development of related tasks such as trajectory prediction,robot control,system certification,etc.,we propose a Bi-GRU neural network based biometric framework.It can effectively recognize the gesture type by analyzing hitters' joint movement information.In addition,by using the unique information from different batters' behavioral habits,we also explored the feasibility of dynamic gesture based batter identification.The experiments proves thatthe dynamic gesture based hitter identification is indeed feasible,and in the gesture data set we collected from 60 individuals,the proposed biometric framework acquired a 100% gesture recognition rate and a 96.96% hitter identification rate,respectively.
Keywords/Search Tags:Table tennis robot, Trajectory prediction, Stereoscopic vision, Gesture based user identification, Table tennis recognition
PDF Full Text Request
Related items