Research On 3D Trajectory Prediction Of Table Tennis Based On Multi Vision

The research on the space trajectory of high-speed moving flying objects currently has very important research significance and application value in the fields of sports,military,aerospace and industry.Table tennis has the characteristics of small size,fast flight speed and complex motion model,which is very suitable for the experimental object of flying object trajectory research.This paper takes table tennis as the research object to carry out the research of three-dimensional trajectory prediction of flying objects.A trajectory extraction system based on multi vision and a trajectory prediction system based on neural network deviation correction of simple physical motion model are built,which is verified by experiments in both simulation environment and real scene.The main research contents and contribution packages of this paper are as follows:(1)Aiming at the research on the trajectory extraction of flying table tennis,three highspeed industrial cameras are used to build a table tennis trajectory extraction system based on multi vision.Its research content mainly includes the principle analysis of multi camera vision,multi view image stereo matching,multi camera calibration and multi camera information fusion.For the stereo matching problem in multi view images of table tennis,in order to solve the problem that the traditional stereo matching algorithm can not meet the real-time and accuracy at the same time,the neural network target recognition method is used to extract the pixel coordinate value of the center point of the target frame for pixel matching,which mainly includes image acquisition,data set production and neural network model training.For the problem of multi camera calibration,Zhang Zhengyou calibration method is used to calibrate the internal and external parameters of multi camera.The calibration experiment is carried out by using the dual target toolbox in MATLAB.In order to improve the tracking accuracy of trajectory extraction system,a multi camera information fusion method based on dynamic weight is proposed according to the mathematical principle and its redundant information characteristicsof multi vision.Experiments show that the proposed method can greatly improve the tracking accuracy of trajectory extraction system.(2)In order to solve the problem that some model parameters are difficult to measure and the model is too complex in the traditional physical motion model of table tennis trajectory,a trajectory prediction model(SPM-LSTM)is proposed,which combines the constraints of simple physical motion model and double LSTM neural network deviation correction.By building a simple physical motion model(SPM),the physical constraint relationship of table tennis trajectory is constructed.The double LSTM neural network is used to train the deviation data between the predicted value of physical model and the real trajectory value iteratively.The trained double LSTM model is used to predict the trajectory deviation value in real time.Finally,the modified predicted trajectory of table tennis is obtained.The simulation comparison experiment proved the accuracy of the SPM-LSTM model trajectory prediction,and the real scene experiment proved that the SPM-LSTM model can predict the low-speed table tennis trajectory by the robot and achieve a certain success rate of hitting.