基于双目视觉的手势关节点跟踪方法  

Method for Hand Joint Tracking Based on Binocular Vision

作　　者：高璐 史金龙[1] 孔秀 GAO Lu;SHI Jinlong;KONG Xiu(School of Computing,Jiangsu University of Science and Technology,Zhenjiang 212003)

机构地区：[1]江苏科技大学计算机学院,镇江212003

出　　处：《计算机与数字工程》2020年第6期1344-1349,共6页Computer & Digital Engineering

基　　金：国家重点研发计划(编号:2018YFC0309100,2018YFC0309104);江苏省六大人才高峰专项(编号:1612991602)资助。

摘　　要：随着交互方式的发展,传统的鼠标、键盘已经不能满足用户的需求;基于视觉的手势识别和跟踪可以更加方便快捷地实现用户愿景,大幅度增强用户体验。为此,论文对手势跟踪进行研究,模拟人体双目视觉,从二维视频中提取手势的运动参数。运用基于仿射变换的马氏距离算法和基于金字塔光流的动态线性模型法进行特征点匹配和时间序列跟踪,从而有效提高了精确度。另外,根据运动模型可以进行轨迹预测,在动态手势跟踪和人机交互等领域具有一定的研究和应用价值。With the development of interactive methods,traditional mouse and keyboard can no longer meet the needs of users. Vision-based gesture recognition and tracking can more easily and quickly achieve the user’s vision and greatly enhance the user experience. To this end,the paper studies the gesture tracking by simulating the binocular vision of the human body and extracting the motion parameters of the gesture from the 2 D video. The feature point matching and time series tracking are performed by the Markov distance algorithm based on affine transformation and the dynamic linear model method based on pyramid optical flow,which effectively improves the accuracy. In addition,trajectory prediction can be performed according to the motion model,and it has certain research and application value in the fields of dynamic gesture tracking and human-computer interaction.

关 键 词：手势跟踪 运动模型 特征点匹配 轨迹预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]