Hand gesture recognition for HCI

Abstract

Gesture driven input is becoming ever more useful due to the growing market of touch enabled smartphones. However the same cannot be said for desktop and notebook computers, through which most of our digital creations and ideas are built and presented. This has lead researchers to rethink the means through which we interact with desktop and laptop computers. This document presents a hand gesture recognition system which enables an additional vision based layer, without attempting to replace the keyboard and mouse, but rather to accompany it just like gesticulations accompanies our every day speech. A dynamic feature driven system was built as a proof of concept to illustrate the benefits of such a system. It acquires an image from a standard webcam, isolates the skin colour to reduce the search space, trains by storing templates and uses basic template matching to recognize hand gestures. Experimental results show that, under a complex background lit only by artificial white light, the system is capable of achieving a recognition rate of 98% with 16 templates at 17 frames per second using pixel template matching.