Simple and flexible vision-based hand gesture recognition system.

Abstract

Rapid development in Human Computer Interactions (HCI) has created various alternatives to provide convenient communication channels between human being and computers. The traditional communication devices such as keyboard and mouse have proven to be cumbersome for real time applications in 3D environment. A more natural and flexible hand gesture recognition is therefore considered as a suitable alternative since no external equipment is to be worn by the operators. This paper proposes a simple yet flexible Hand Gesture Recognition System-Codebook (HGRS-Codebook) using the techniques of codebook background subtraction, skin color segmentation, median filtering, contour detection, convexity defects and definition of peaks and valleys of fingertips. The fundamental concept of the system is to count the number of fingers displayed in particular gesture based on the unique characteristic of human hands which normally constituted by five fingers in a circle-like palm. The developed system is capable of recognizing six gestures ranging from “zero” to “five”. It is invariant in translation, rotation and scale due to its nature in gesture classification. In addition, it is user independent and is able to deal with different backgrounds and lighting conditions. HGRS-Codebook achieved recognition rate averaged at 90.33% for indoor environment and 80% for outdoor environment.