Research On Human Motion Attitude And Position Estimation Technology Based On Sparse MEMS

Human motion posture and position estimation technology can analyze human motion based on wearable sensor data,and reconstruct human motion posture and position.It has a wide range of application values in movies,games,medical care,sports and other industries.In this thesis,aiming at the problems of too many wearable sensors and large errors in motion posture and position estimation in the current inertial micro-electromechanical system(Micro Electro Mechanical System,MEMS)human motion posture and position estimation technology,this thesis focuses on the sparse MEMS sensor posture calculation.,Human skeleton posture estimation,human root displacement estimation and other algorithms,design and implement sparse MEMS human motion posture and position estimation application system.The main innovations are as follows:1.A sparse MEMS human pose estimation algorithm based on temporal convolutional network and bone position correction is proposed.Based on the temporal convolution network to learn the temporal and spatial correlation characteristics of human skeleton motion,the algorithm effectively improves the performance of human skeleton pose estimation while reducing the number of MEMS.2.A human root displacement estimation algorithm based on biped zero-velocity interval acceleration integral correction and bone position conduction is proposed.Through the zero-velocity interval detection of the foot and bone position transmission,the human root node displacement is estimated in real time,which reduces the cumulative error of the human root displacement estimation,improves the action adaptability of the root displacement estimation,and effectively improves the estimation performance of the human root displacement.3.An algorithm for constructing a dataset of human action poses based on a virtual character model is proposed.By collecting the action data of the virtual character model instead of the real human body,it solves the problem of difficult large-scale data collection of real human body movement postures,enriches the data of human body movement postures,and improves the performance of human movement prediction.This thesis designs and completes the application system of human motion posture and position estimation.While realizing the high-precision and low-latency estimation of human motion posture and position,it drives the 3D virtual character model to move synchronously based on Unreal Engine(UE).Finally,this thesis builds a test system to test and verify the key technologies.The results show that: in terms of function realization,the reconstruction task of human motion posture and position is realized,and the number of sensors is reduced from 13 to 6;In terms of performance,the average direction error is reduced by 13.59% compared with the current cycle neural network algorithm;in terms of root displacement estimation performance,the closure error is reduced by 77.49%compared with the traditional zero-speed update algorithm.To sum up,the research results of this thesis can better solve the main problems in MEMS human motion attitude and position estimation technology.