Computer Vision Based Realtime Fingertip Detection for Online Symbol Recognition
The use of keyboard and mouse introduced with the advent of the first GUI operating system is a proven way of interaction with interfaces based on WIMP (window, icon, menu and pointer) paradigm, which has prevailed for few decades. Traditional input devices have limitations as they are originally designed to support two dimensional interfaces in single user environment. Lately there is increased interest on novel ways of interaction as future interfaces demand new interaction paradigms to be developed. Hence, research on adopting natural interaction for human computer interaction has gained increasing attraction and it promises to eliminate cumbersome intermediate input devices. This project, aims to develop an alternative interaction framework for natural interaction with gesture recognition using computer vision and online symbol recognition. The framework uses commonly available hardware such as webcam and cell phone. It uses real time hand gesture recognition to replicate mouse interface and makes a unique attempt to use symbol recognition to execute keyboard shortcuts. The symbol recognition algorithm implemented using Java follows a statistical approach and additionally, it was extended to a Java enabled phone. OpenCV was used for image processing. Tests prove that the approach used for fingertip detection allows person independent skin color detection. The combined accuracy rate for symbol recognition on both desktop and cell phone accounts for 90.1％ and evaluation of the framework with end-users and experts confirmed that the framework is easy to adapt and most evaluators commended on the richness of interaction and usability of the framework.
computer vision gesture recognition human computer interaction ubiquitous
M.H.M. Hamzeen M.U.S. Perera
Department of Computing Informatics Institute of Technology Colombo, Sri Lanka
国际会议
2010 International Conference on Software and Computing Technology(2010年软件与计算机技术国际会议 ICSCT 2010)
昆明
英文
439-443
2010-10-17（万方平台首次上网日期，不代表论文的发表时间）