Advanced Gesture and Pose Recognition Algorithms Using Computational Intelligence and Microsoft KINECT Sensor

This research work suggests one kind of approach in developing a natural human-robot interface that will be used for control of four wheels differentially steered mobile robot. The designed system is capable of extracting, understanding and learning a sequence of full body gestures and poses, that were previously captured in standard RGB and IC DEPTH videos. The starting set of robot commands in first study case, includes the following 4 postures: START, WAIT A MINUTE, STOP and SLOW DOWN, while in the second study case 5 gestures were realized: START, TURN RIGHT, TURN LEFT, SPEED UP and SLOW, while command STOP is realized as pose. The special feature of proposed classifier system is fact that human user is always in visual domain of camera but without fixation for defined position or orientation. Two different kinds of classifiers were implemented: first, support vector machine (SVM) classifier and second, based on multiple interconnected FUZZY Logic systems. The research showed satisfactory results in small classification error, simple human operator training and user comfort.