The usage of hidden Markov models in the vision system of a mobile robot

This paper presents the development and experimental implementation of remote control of mobile robots using visually recognized bask hand gestures presented as input commands. The hand is segmented from the background by taking advantage of the unique coloration of human skin. Color histogramming techniques and Gaussian estimation are applied at the image preprocessing stage. To shorten the sampling period the Kalman filter and technique based on position prediction of the bounding box containing the hand are applied. The recognition module uses hidden Markov models. Some practical experiments conducted in laboratory are briefly discussed Preliminary results obtained are promising and show that, on the basis of the software developed as well as the Khepera robot equipped with a relatively cheap color camera, we can conduct several advanced experiments in naturally occurring circumstances in our laboratory.