Robust implementation of hand gesture recognition for remote human-machine interaction

Abstract

A robust hand gesture recognition algorithm for remote human-machine interaction is proposed that has been optimized for implementation on an embedded platform. Hue-saturation-value (HSV) thresholding and unit-gradient vector (UGV) background subtraction methods are employed to overcome common issues related to variations in lighting conditions. Top-hat transformation is used to detect fingers and hand gestures, which are translated to command inputs for remotely controlling a media player. Experimental results demonstrate that the algorithm performs efficiently and accurately on an embedded board with an average computational cost of 143 millisecond per gesture and is robust to changes in illumination.