Towards Naturalistic Interfaces of Virtual Reality Systems

Interaction plays a key role in achieving realistic experience in virtual reality (VR). Its realization depends on interpreting the intents of human motions to give inputs to VR systems. Thus, understanding human motion from the computational perspective is essential to the design of naturalistic interfaces for VR.

This dissertation studied three types of human motions, including locomotion (walking), head motion and hand motion in the context of VR.

For locomotion, the dissertation presented a machine learning approach for developing a mechanical repositioning technique based on a 1-D treadmill for interacting with a unique new large-scale projective display, called the Wide-Field Immersive Stereoscopic Environment (WISE). The usability of the proposed approach was assessed through a novel user study that asked participants to pursue a rolling ball at variable speed in a virtual scene. In addition, the dissertation studied the role of stereopsis in avoiding virtual obstacles while walking by asking participants to step over obstacles and gaps under both stereoscopic and non-stereoscopic viewing conditions in VR experiments.

In terms of head motion, the dissertation presented a head gesture interface for interaction in VR that recognizes real-time head gestures on head-mounted displays (HMDs) using Cascaded Hidden Markov Models. Two experiments were conducted to evaluate the proposed approach. The first assessed its offline classification performance while the second estimated the latency of the algorithm to recognize head gestures. The dissertation also conducted a user study that investigated the effects of visual and control latency on teleoperation of a quadcopter using head motion tracked by a head-mounted display. As part of the study, a method for objectively estimating the end-to-end latency in HMDs was presented.

For hand motion, the dissertation presented an approach that recognizes dynamic hand gestures to implement a hand gesture interface for VR based on a static head gesture recognition algorithm. The proposed algorithm was evaluated offline in terms of its classification performance. A user study was conducted to compare the performance and the usability of the head gesture interface, the hand gesture interface and a conventional gamepad interface for answering Yes/No questions in VR.

Overall, the dissertation has two main contributions towards the improvement of naturalism of interaction in VR systems. Firstly, the interaction techniques presented in the dissertation can be directly integrated into existing VR systems offering more choices for interaction to end users of VR technology. Secondly, the results of the user studies of the presented VR interfaces in the dissertation also serve as guidelines to VR researchers and engineers for designing future VR systems.