LPI: learn postures for interactions

To a great extent, immersion of a virtual environment (VE) depends on the naturalness of the interface it provides for interaction. As people commonly exploit gestures during communication, therefore interaction based on hand-postures enhances the degree of realism of a VE. However, the choice of selecting hand postures for interaction varies from person to person. Generalizing the use of a specific posture with a particular interaction requires considerable computation which in turns depletes intuition of a 3D interface. By investigating machine learning in the domain of virtual reality (VR), this paper presents an open posture-based approach for 3D interaction. The technique is user-independent and relies neither on the size and color of hand nor on the distance between camera and posing-position. The system works in two phases—in the first phase, hand-postures are learnt, whereas in the second phase the known postures are used to perform interaction. With an ordinary camera, a scanned image is partitioned into equal size non-overlapping tiles. Four light-weight features, based on binary histogram and invariant moments, are calculated for each part and portion of a posture-image. The support vector machine classifier is trained by posture-specific knowledge carried accumulatively in each tile. By posing any known posture, the system extracts the tiles information to detect a particular hand-posture. At the successful recognition, appropriate interaction is activated in the designed VE. The proposed system is implemented in a case-study application; vision-based open posture interaction using the libraries of OpenCV and OpenGL. The system is assessed in three separate evaluation sessions. Results of the evaluations testify efficacy of the approach in various VR applications.