Kinect人体运动捕捉误差及其空间分布  

Positioning error and its spatial distribution of motion capture with Kinect

作　　者：吴玮莹 王豫[1] 李秦[1] 攸兴杰 陈天远 蒲放[1] 樊瑜波[1] 

机构地区：[1]北京航空航天大学生物与医学工程学院,北京100191

出　　处：《北京生物医学工程》2014年第4期344-348,共5页Beijing Biomedical Engineering

基　　金：国家自然科学基金(61201313)资助

摘　　要：目的针对微软发布的体感捕捉设备Kinect开展精度测试实验,检测系统定位的位置误差及其空间分布,为基于Kinect体感交互的生物医学工程应用提供参考。方法本实验利用Kinect和高精度运动捕捉设备NDI Optotrak同时进行人体运动捕捉,并采用Kinect和NDI进行测量,结果取相对位移值进行比较。以NDI的测量数据作为真实数据的有效近似,评估Kinect的测量精度。结果本实验最终测得Kinect识别精度误差为(0.0 283±0.0 186)m,均方根误差为0.0 303 m,水平角度识别范围为51.49°。结论 Kinect的精度和稳定性在cm级,可应用于康复训练、手术室设备控制等生物医学工程领域,但对于精度要求更高的领域,如智能手术机器人的控制等,其精度和稳定性还有待提高。Objective To achieve the accuracy measurement by using the human-computer natural technique of the somatosensory capture device Kinect and to provide reference for the application prospect of the human-computer natural interaction in biomedical engineering （BME） field. Methods In this experiment, Kinect is used to locate the space coordinates and compared with the NDI motion capture system. The position is located at each measurement point both by Kinect and NDI. The relative displacement is adopted to assess the accuracy of Kinect while the data from NDI is used as a valid approximation of the real data. Results In the software and hard ware testing platform we constructed, the error is （0. 0283 ± 0. 0186） m, and the root-meansquare error （RMS） is 0. 0303m. These data are collected over a range of 51.49 degrees. Conclusions The accuracy and stability of Kinect are both in centimeter degree. It is acceptable in certain BME applications like rehabilitation training and the control of the operating room devices. The accuracy and stability need to be improved in other fields with higher requests, like the control of operating room robots.

关 键 词：KINECT 自然交互 手势识别 精度模型 

分 类 号：R318.6[医药卫生—生物医学工程]