惯导/视觉导航/人体运动特征信息融合模型与高精度人体动作捕获方法研究

The arrival of information and intelligent era make the human movement to capture and behavior understanding has been more and more applications. The capture and identify high-precision action of human body is challenging research topics, and the bionic engineering, medical engineering, sports and film animation and other industries to promote the important significance. In order to realize the high precision positioning, attitude measurement and motion capture of the human body, this project first studies an INS/visual integrated navigation method based on road sign recognition, compensates the INS error according to the position information provided by road signs. Then, the method of combining the information of human motion characteristics is studied, and the INS error is corrected by using human motion zero speed, attitude information and human joint constraint information, and the human body action recognition is carried out according to the calculated attitude information. Based on the information of the visual information and the characteristics of the human body, the fusion model of the INS/human characteristics navigation system is established, and the optimal estimation and compensation of the errors are realized based on Kalman filter. Finally, human body positioning and motion recognition experiments are designed, according to the positioning accuracy and motion identification accuracy the effectiveness of the proposed INS/visual navigation/human characteristics of multi-information source fusion model and high-precision action recognition method is verified.

信息化与智能化时代的到来使得人体动作捕获和行为理解得到了越来越多的应用，人体高精度动作捕获与识别是富有挑战性的研究课题，对仿生工程、医学工程、体育竞技以及电影动画等产业的推动具有重要的意义。为实现人体的高精度定位、姿态测量与动作捕获，本项目首先研究一种基于路标识别的惯导/视觉组合导航方法，根据路标提供的位置信息对惯导的误差进行补偿，实现室内人体高精度自主定位。然后，研究一种结合人体运动特征的信息的组合方法，利用人体运动零速、姿态特征信息、人体关节点约束信息等对惯导误差进行修正，并根据解算出的姿态信息进行人体动作识别。结合视觉信息和人体特征信息建立惯导/视觉导航/人体特征多信息源融合模型，基于卡尔曼滤波实现各项误差的最优估计与补偿。最后，设计人体定位与动作识别试验，根据定位精度和动作识别准确度验证提出的惯导/视觉导航/人体特征多信息源融合模型与高精度动作识别方法的有效性。

信息化与智能化时代的到来使得人体动作捕获和行为理解得到了越来越多的应用，人体高精度动作捕获与识别是富有挑战性的研究课题，对仿生工程、医学工程、体育竞技以及电影动画等产业的推动具有重要的意义。为实现人体的高精度定位、姿态测量与动作捕获，本项目开展了基于路标识别的惯导/视觉组合定位方法研究、人体动作捕获总体方案研究、基于互补滤波的人体动作姿态测量方法研究、单肢体动作姿态测量实验研究以及手臂动作姿态测量及重构研究、基于卡尔曼滤波的单肢体动作姿态测量方法研究、多肢体动作姿态测量方法研究及人体动作姿态重构研究。本项目建立了一种基于路标的视觉导航模型，并基于此提出了一种MEMS惯导/视觉导航组合高精度定位方法；设计了一种人体高精度动作姿态测量与捕获的总体方案；提出了一种基于互补滤波的人体动作姿态测量方法，并进行了实验验证，结果显示肢体姿态测量误差小于2.2°；提出了一种基于卡尔曼滤波的人体动作姿态测量方法，并进行了实验验证，结果显示肢体姿态测量误差小于2°。本项目研究成果可以为人体与智能载体动作识别、自主定位等重要领域提供方法和技术支持。