基于惯性传感器的人体运动姿态三维重构

发布时间：2018-05-19 13:51

本文选题：惯性传感器 + 虚拟现实　；参考：《哈尔滨工程大学》2013年硕士论文

【摘要】：高效、便捷地获取人体运动姿态信息在动漫制作、游戏开发、影视制作等众多领域具有重要的应用价值。目前已经取得应用成果的人体运动捕捉重构方式主要有机械式、声学式、电磁式和光学式等，这些方式大多存在成本高、对动作限制大和使用不方便等缺点。本文针对以上这些常见的运动捕捉重构方式所存在的缺点，研究了可用于惯性传感器的人体运动捕捉技术中人体运动模型建立的一些基本方法，设计了基于刚体动力学的三维层次人体骨骼模型，采用基于惯性传感器的动作捕捉方式。这种方式具有操作方式简单、价格低廉、对环境要求低和对运动动作限制小的优点。根据人体结构和运动人体科学理论知识，本文设计了惯性传感器可以在人体运动肢体上以任意姿态朝向绑定的姿态初始化标定方法。通过初始化标定时的特定动作可以实现对传感器和关节的对应识别，并计算出各个惯性传感器的自身坐标系与三维人体骨骼模型中对应的人体骨骼坐标系的坐标转换关系。然后通过各个惯性传感器姿态坐标系的坐标转换和对人体骨骼树进行深度优先遍历计算，，完成人体运动肢体姿态数据的实时更新，驱动人体骨骼模型模拟人体动作，实现了对动作的实时捕获跟踪。在计算机三维虚拟场景中重构人体运动姿态需要绘制三维人体模型，为了使得整个动作捕捉系统具有更为广泛的应用前景，本文选择了微软公司提供的适用于Windows平台的Direct3D图形绘制接口。通过X文件加载网格模型数据中的骨骼蒙皮信息和骨骼层次信息，在进行动作跟踪时根据惯性传感器捕获到的骨骼姿态角数据实时更新人体模型中的骨骼变换矩阵信息，以此确定人体各个骨骼的位置和朝向，实现了三维人体骨骼蒙皮动画。为了能够最大限度地减少动作捕捉装置对人体运动动作的限制，本文采用了无线数据传输方式，并制定了惯性传感器与上位机之间的数据通信格式。为了能够方便地对大量运动数据进行统计和分析，本文结合目前软件开发常用的MySQL开源数据库，设计了人体运动姿态数据库。将采集到的实时人体姿态数据存入数据库后，可进一步用于人体健康、生物力学、动作预测和行为识别等科学研究。各个惯性传感器采集到人体运动肢体姿态数据后将数据传输到PC终端机，PC终端机上的上位机软件实现了对人体运动姿态的三维重构。从三维重构结果来看，人体运动动作跟踪实时性良好，动作重构效果逼真。
[Abstract]:It has important application value in many fields such as animation production, game development, film production and so on. At present, the main methods of human motion capture and reconstruction have been obtained, such as mechanical, acoustic, electromagnetic and optical. Most of these methods have the disadvantages of high cost, large restrictions on movement and inconvenient use. In view of the shortcomings of these common motion capture and reconstruction methods mentioned above, this paper studies some basic methods for the establishment of human motion model, which can be used in human motion capture technology for inertial sensors. A three-dimensional hierarchical human skeleton model based on rigid body dynamics was designed, and the motion capture method based on inertial sensor was adopted. This method has the advantages of simple operation, low price, low requirement to environment and small restriction on movement. Based on the scientific knowledge of human body structure and motion human body, this paper designs an initialized calibration method in which the inertial sensor can bind in any attitude direction on the human body moving limb. By initializing the specific action of calibration, the corresponding recognition of sensors and joints can be realized, and the coordinate transformation relationship between each inertial sensor's own coordinate system and the corresponding human skeleton coordinate system in three-dimensional human skeleton model can be calculated. Then through the coordinate transformation of each inertial sensor attitude coordinate system and the depth first traversal calculation of the human skeleton tree, the real-time update of the human body motion limb posture data is completed, and the human skeleton model is driven to simulate the human body action. The real-time capture and tracking of the action is realized. In order to make the whole motion capture system have a more extensive application prospect, it is necessary to draw a three-dimensional human body model in order to reconstruct the human motion posture in the computer 3D virtual scene. In this paper, we choose the Direct3D graphic drawing interface for Windows platform provided by Microsoft. By loading the skeleton skin information and skeleton level information from the mesh model data in the X file, the skeleton transformation matrix information in the human body model can be updated in real time according to the skeletal attitude angle data captured by the inertial sensor in the course of motion tracking. In order to determine the position and orientation of each human skeleton, a three-dimensional human skeleton skin animation is realized. In order to minimize the limitation of motion capture device on human motion, this paper adopts wireless data transmission mode, and formulates the data communication format between inertial sensor and upper computer. In order to make statistics and analysis of a large number of motion data conveniently, this paper designs a human motion posture database based on MySQL open source database, which is commonly used in software development. The collected real-time human posture data can be used in human health, biomechanics, motion prediction and behavior recognition. After each inertial sensor collects the human body motion limb posture data, the data is transmitted to the PC terminal PC terminal computer software to realize the 3D reconstruction of the human body motion posture. From the result of three-dimensional reconstruction, the real-time motion tracking of human body is good, and the effect of motion reconstruction is lifelike.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.9;TP212

【参考文献】