Gesture Recognition from Two-Person Interactions Using Ensemble Decision Tree

The evolution of depth sensors has furnished a new horizon for human–computer interaction. An efficient two-person interaction detection system is proposed for an improved human−computer interaction using Kinect sensor. This device is able to identify twenty body joint coordinates in 3D space among which sixteen joints are selected and those have been adapted with certain weights to form four average points. The direction cosines of these four average points are evaluated followed by the angles made by x, y and z axes, respectively, i.e., twelve angles have been constructed for each frame. For recognition purpose, ensemble of tree classifiers with bagging mechanism is used. This novel work is widely acceptable for various gesture-based computer appliances and yields a recognition rate of 87.15%.