Font Size: a A A

Human Motion Posture Attitude Estimation And Recognition Based On Deep Natural Network

Posted on:2021-05-31Degree:MasterType:Thesis
Country:ChinaCandidate:Y WangFull Text:PDF
GTID:2428330620964243Subject:(degree of mechanical engineering)
Abstract/Summary:PDF Full Text Request
With the development of deep neural network,there is a great breakthrough in the field of human posture estimation and recognition.Through the deep neural network,the low-level feature information of the image can be combined to estimate and recognize the human posture in a higher level.This method reduces the need of external devices in the field of human pose estimation and recognition,and provides a theoretical basis for human pose evaluation system.In this paper,a human body posture evaluation system is designed based on the estimation and recognition of badminton players' movement posture,which is used to estimate and recognize the human body posture in the process of badminton players' movement.This paper summarizes the differences and connections between attitude estimation and motion recognition,as well as the advantages and disadvantages of estimating human posture coordinates in different dimensions.Through the analysis of badminton players' movement posture,three depth neural network models are reproduced.The advantages and disadvantages of each model are verified by using badminton players' image set,and the basic model is selected to carry out the follow-up design of human posture evaluation system.In the design process,firstly,the best angle for estimating the human posture of badminton players is given by analyzing the camera placement angle in the actual field;secondly,the best camera parameters are given by analyzing the camera parameters and combining with the actual engineering data set and verified by the human posture evaluation formula;secondly,the compression operation and interval sampling are used to improve the estimation speed;finally,it is discussed On the problems and solutions of human posture evaluation,and through experiments to compare the results.In this paper,a new depth neural network model is proposed to estimate the human body posture of badminton players in a single frame image.It has good adaptability in estimating the related posture of badminton players and good performance in estimating accuracy and speed.The traditional human body posture model is optimized by combining the modification of badminton movement characteristics.The estimated human body posture coordinate system is used to Finally,the software and hardware design of the whole system are realized,and the effect of the estimation model,the human posture model and the evaluation algorithm are verified by the actual test.
Keywords/Search Tags:human posture estimation and recognition, badminton, depth neural network, local evaluation, similarity
PDF Full Text Request
Related items