Conclusion and Future Work

For a given task and work environment, end-effector positions will be similar for a wide range of body sizes. Our system is designed to take advantage of this similarity, by measuring the absolute 3D cartesian space coordinates of the points of interest on the body, and using them directly as goals for end-effector constraints, rather than measuring or deriving the operator's joint angles. Thus, while the model's posture may not precisely match the operator's, the end effectors of the constraints are always correct. This is very important, especially in situations where the operator is controlling a human model of different size in a simulated environment. Additionally, by obtaining our data in a relatively unobstructive manner, we insure that the operator moves in a natural way, unencumbered by measuring equipment.

We have not run experiments to determine the absolute accuracy or fidelity of our tracking system between the recreated posture and the actual posture of the operator. With the collision avoidance enabled, the generated postures are generally accurate, if not exact, representations of the actual posture. For high-fidelity motion recording, this is not good enough, but for most interactive applications it is an extremely useful technique.

Some future enhancements:

• We will use our anthropometric scaling system (SASS) to better choose a correct body size to match that of the operator. That is a matter of interactively anthropometrically scaling (as opposed to regular scaling) the human figure from 5th percentile up to 95th percentile, stopping where the sensor locations fit best [3][1].

• Another problem present is sensor noise. We need better noise filtering algorithms, both for noise from the actual equipment, and noise from the human operator. Humans tend to twitch and oscillate at certain frequencies, and these could be filtered out.

• Our simulation will be augmented to allow the participant to interact more directly with the environment. Actions can be initiated with direct actions (e.g. press a button to grasp), or indirectly; for example, placing both hands in intersection with an object can be a cue for grasping [2]. Sound rendering will be used to add realism and to signal events, such as collisions or warnings [10].

Even at this stage of development, the virtual Jack model is proving to be a useful tool in fast, intuitive input of whole body posture and location with minimal sensing hardware. Our system provides a method of interacting with a virtual environment without encumbering the participant.