基于经验的抓取姿态构造及数据手套校正

针对数据手套校正过程中真实人手姿态获取困难、校正模型简单的问题,提出一种基于真实抓取经验的校正数据库构造方法.该算法根据人手生理结构和手套传感器布局建立具有更高运动逼真度的虚拟手运动学模型,实现更为灵活的拇指运动和软手掌效果;在不需要任何外部硬件设备的情况下,将真实抓取经验与最优化理论结合,寻找针对给定三维物体及操作任务的最优抓取姿态,从而构造真实人手姿态数据库;利用最小二乘拟合和闭环迭代算法获得精确的数据手套传感器输出模型.实验结果表明:所提出的姿态构造方法能有效提高数据手套校正效率,且操作过程简单,易于实现自动化校正,适合实际工程应用.

Experience based grasping gesture synthesis and data glove calibration

An experience based gesture database construction method is proposed to solve the two primary problems for current calibration routine as: the difficulty for ground-truth data gathering and the oversimplified calibration model. By considering the anatomic structure of human hand and sensor layout of the data glove, the method established an accurate human hand kinematics model, which can realize flexible thumb movement and soft palm effect. Given the 3D object to grasp and the desired grasp task, an optimization procedure, combined with daily experience for grasping, was proposed for hand gesture synthesis. Least square regression and closed kinematics theory were used to find the accurate sensors’output transform parameters. The work described provides a novel method for ground-truth database construction. This approach has the advantages that it does not require any special outer device and can be implemented automatically, while still produce high fidelity performance.