Computer Vision Based Gesture Recognition for Desktop Object Manipulation

This paper proposes a Kinect based real time gesture recognition system to manipulate desktop object. This system can be divided into three subsystems. 3D position of hand is detected by the depth sensor of Kinect. These acquired position point are then analyzed to recognize pre-defined gestures. Recognized gestures are then implemented on a desktop to manipulate different objects. Training is done with the help of implementing a machine learning technique known as Hidden Markov Model (HMM). The HMM was trained using Baum-Welch algorithm. Satisfactory gesture recognition has been achieved using this model. More than 1200 gestures were used to train the system, which yielded an accuracy of 89%.