Real-Time Marker-Based Finger Tracking with Neural Networks

Abstract

Hands in virtual reality applications represent our primary means for interacting with the environment. Although marker-based motion capture with inverse kinematics (IK) works for body tracking, it is less reliable for fingers often occluded when captured with cameras. Many computer vision and virtual reality applications circumvent the problem by using an additional system (e.g. inertial trackers). We explore an alternative solution that tracks hands and fingers using solely a motion capture system based on cameras and active markers with machine learning techniques. Our animation of fingers is performed by a predictive model based on neural networks, which is trained on a movements dataset acquired from several subjects with a complementary capture system (inertial). The system is as efficient as a traditional IK algorithm, provides a natural reconstruction of postures, and handles occlusions.