Approach
To test filtering options, we have implemented real-time filtering software (Saul) that handles the flow of marker data from the Vicon motion capture system to the Lithtech engine. Before sending the data to Lithtech, Saul first applies a sequence of C++ filter classes we've written to it. So far, we have tested low-pass and Kalman filters.

Next Steps
Based on our first test, we plan to experiment more with the parameters of the Kalman and low-pass filters and create a system that allows us to apply fine-tuned parameter sets to different body segments simultaneously.

learn more about filtering

Results
Low-pass   [video]
- Relatively easy to implement and are good to 'learn on'
- Not good for quick motions
- Smooths too much - characters skate

Kalman   [video]
- Can't handle lost frames, popping due to ghost points and mislabeling, and discontinuous trajectories
- Ability to set parameters adds complexity but much greater control
- Requires complicated quaternion calculations
- Because Kalman filters are predictive, different velocities of body parts caused severe stretching

   [video] No Filter