基于视觉的手势界面关键技术研究  

Research on Key Issues of Vision-Based Gesture Interfaces

作　　者：武汇岳[1,2] 张凤军[1] 刘玉进[1] 戴国忠[1] 

机构地区：[1]中国科学院软件研究所人机交互技术与智能信息处理实验室,北京100190 [2]中国科学院研究生院,北京100049

出　　处：《计算机学报》2009年第10期2030-2041,共12页Chinese Journal of Computers

基　　金：国家"九七三"重点基础研究发展规划项目基金(2009CB320804);国家自然科学基金(U0735004;60673188);国家"八六三"高技术研究发展计划项目基金(2009AA01Z337;2008AA01Z303)资助

摘　　要：针对视觉手势界面存在的问题,提出了一套行之有效的解决方案.首先,为了解决视觉手势交互中的Midas Touch问题,以人类注意的信息加工模型为理论依据提出了一个可扩展的视觉手势交互模型,该模型将手势交互过程分为选择性处理、分配性处理和集中处理3个不同阶段;然后,基于该模型提出了一个视觉手势识别框架,并结合认知心理学从手势检测、跟踪和识别3个方面对该框架的各个组成模块的关键技术进行了阐述,其中手势检测模块和识别管理模块能够辅助系统在复杂的背景中滤除掉不相关信息而选择性地搜索人手并根据上下文信息对手势识别任务重定向,从而避免了系统时刻都处于激活状态并对所有的手势动作都进行识别分析,有效解决了Midas Touch问题.文中介绍了使用该方法实现的IEToolkit手势界面工具平台,并基于一个视觉手势交互系统进行了实验测试与评估,结果验证了文中方法的可用性.In this paper, an effective solution is described for vision-based gesture interfaces, which can help avoid what is often called the Midas Touch Problem, where everything the user does is interpreted as an interaction. First, an interactive model for hand gesture is presented based on information processing model of human attention, which divides gesture interaction into selective process, divided process and sustained process; Then a recognition framework is pro- posed based on the interactive model and is interpreted from the perspective of cognitive psychol- ogy. In this framework, the hand detection model and the recognition management model can help recognize hand gesture from complex environment and analyze specific context and redirect gesture recognition into the adequate system model, thus minimizing the need to activate all the different gesture recognition types simultaneously and solving the Midas Touch Problem effec- tively. At the end, a vision-based gesture interfaces toolkit for interactive games is presented based on the recognition framework. This paper also presents experimental results based on an interactive prototype, regarding the speed, accuracy and robustness of the implemented system, which validate the quality and usability of the proposed.

关 键 词：人机交互 视觉手势界面 交互模型 识别框架 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]