基于Kinect和ODE的人体运动仿真

发布时间：2018-01-21 07:41

本文关键词： 仿人机器人运动建模人体运动仿真动作评价　出处：《沈阳航空航天大学》2017年硕士论文　论文类型：学位论文

【摘要】：仿人机器人是模仿人类身体结构和功能的机器人,在机器人领域一直是众多研究人员的关注热点。近年来,机器从以往只能被动接受人类操控逐渐向主动学习人类特征的方向过渡,通过示教学习的方式操控并训练机器逐渐成为机器学习的一种主要方法。本文建立了一种人体运动模型,并在该模型的基础上,结合Kinect采集到人体运动数据,驱动虚拟仿人机器人运动,为仿人机器人的研制和人体运动的研究提供了实验平台、模型和方法。首先,在实验室原有的基础之上完善了运动仿真平台,该平台集运动数据采集、实时数据分析、实时仿真再现、动作相似度评价等功能于一体。使用ODE在该平台上创建了一个运动学仿真环境,可以在该环境中进行刚体运动仿真学实验,为人体运动建模、仿真提供了平台基础。其次,模仿真实人体结构,创建了一个拥有20个自由度的仿人机器人模型。针对三维人体运动数据的特征,提出了一种基于关节旋转角和旋转轴的人体运动模型,该模型可实时的将Kinect采集到的人体关键点坐标转换为能够驱动虚拟机器人运动的关节角数据,实现了人体运动实时仿真再现功能。最后,为了能够客观的对人体动作做出评价,将两组动作数据的相似度作为评价指标。由于动作数据的长度通常不同,传统的欧式距离往往不能真实的评价两组动作的相似度,因此,引入动态时间规整(Dynamic Time Warping,DTW)算法,将运动数据在时间轴上进行对齐操作,再通过寻找最优规整路径得到两组运动数据的相似程度。仿真实验结果表明本文建立的人体运动模型能够对人类四肢的动作进行实时的仿真再现。动作评价实验结果表明本文提出的动作评价指标能够对人体动作进行较为客观的评价。
[Abstract]:Humanoid robot is a robot that imitates the structure and function of human body. It has been the focus of attention of many researchers in the field of robot. From the past, the machine can only passively accept human manipulation, gradually to the active learning of human characteristics in the direction of transition. The manipulation and training of machines by teaching and learning has gradually become one of the main methods of machine learning. In this paper, a human motion model is established and based on this model. Combined with Kinect to collect human motion data, driving virtual humanoid robot motion, for the development of humanoid robot and the study of human movement provides an experimental platform, models and methods. On the basis of the original laboratory, the motion simulation platform is perfected. The platform collects motion data, real-time data analysis, real-time simulation and reproduction. Using ODE to create a kinematics simulation environment in which rigid body motion simulation experiments can be carried out to model human motion. Simulation provides the foundation of the platform. Secondly, a humanoid robot model with 20 degrees of freedom is created by imitating the real human body structure, aiming at the characteristics of 3D human motion data. This paper presents a human motion model based on the rotation angle and rotation axis of the joint. The model can transform the key points of human body captured by Kinect into the joint angle data which can drive the motion of the virtual robot in real time. Finally, in order to objectively evaluate the human actions, the similarity of the two groups of action data is taken as the evaluation index. Because the length of the action data is usually different. The traditional Euclidean distance can not really evaluate the similarity between the two groups of actions, so the dynamic Time warping (DTW) algorithm is introduced. Align the motion data on the timeline. The simulation results show that the human motion model established in this paper can simulate and reproduce the movements of human limbs in real time. The results show that the motion evaluation index proposed in this paper can evaluate human movement objectively.
【学位授予单位】：沈阳航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】