Quadratic Encoding for Hand Pose Reconstruction from Multi-Touch Input

Abstract

One of the most compelling challenges in virtual reality today is to allow users to carry out virtual manipulation tasks using their hands. Multi-touch devices are an interesting interface for this task, as they are widely available, they provide users with some haptic sensation of their motions, and they give very precise locations of the fingertips. We introduce a quadratic encoding technique to provide plausible and smooth hand reconstructions from multi-touch input at real-time rates suitable for virtual reality applications. Another nice feature of our datadriven approach is that it does not require explicit identification or registration of fingers. We show that quadratic encoding outperforms linear encoding, cubic encoding, and a PCA based inverse kinematics approach, and is well suited for performing real-time virtual manipulation using a multi-touch device.