基于神经网络的非同关节双点缺失值补全研究  

Research on Non Same Joint Double Point Missing Value Completion Based on Neural Networks

作　　者：刘晓曦 丛晓丹 宋昌江 LIU Xiao-xi;CONG Xiao-dan;SONG Chang-jiang(Institute of Intelligent Manufacturing,Heilongjiang Academy of Sciences,Harbin 150090 China)

机构地区：[1]黑龙江省科学院智能制造研究所,黑龙江哈尔滨150090

出　　处：《自动化技术与应用》2025年第2期181-184,共4页Techniques of Automation and Applications

摘　　要：在对运动员的运动姿态进行分析时,需要用动作捕捉系统采集运动员动作,但是在采集过程中,由于摄像头视角原因,并不能覆盖全部实验区域,所以采集到的数据会出现标记点缺失的情况,不方便后期分析运动数据。本文基于Qualisys动作捕捉系统,对运动员滑雪数据进行采集,针对非同关节双标记点缺失的情况,用带有残差结构的全连接神经网络对采集数据进行训练,从而补全摄像头未采集到的缺失标记点,对分析运动姿态具有重要意义。When analyzing the movement posture of athletes,it is necessary to use a motion capture system to capture their movements.However,due to the camera's perspective,the entire experimental area cannot be covered during the collection process,resulting in missing marker points in the collected data,which is inconvenient for later analysis of the movement data.This article is based on the Qualisys motion capture system to collect skiing data from athletes.In response to the situation of missing non same joint dual marker points,a fully connected neural network with residual structure is used to train the collected data,thereby completing the missing marker points that the camera did not collect.This is of great significance for analyzing the movement posture.

关 键 词：动作捕捉系统 神经网络 缺失值补全 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.9[自动化与计算机技术—控制科学与工程]