Rynson Lau: Motion Prediction for Distributed Virtual Environments

Motion Prediction for Distributed Virtual Environments

Although there are some methods proposed for predicting the 3D motion of an object in a distributed virtual environment, most of these methods are primarily designed for predicting the motion of specific objects, by assuming certain object motion behaviors. We notice that in desktop distributed 3D applications, such as virtual walkthrough and computer games, the 2D mouse is still the most popular device being used as navigation input. Through studying the motion behavior of a mouse during 3D navigation, we propose a hybrid motion model for predicting the mouse motion during a 3D walkthrough:

An Elliptic Model for High Velocity Motion: When the mouse is at normal or high velocity motion, we use an elliptic model for predicting future mouse velocity.
A Linear Model for Low Velocity Motion: When the mouse is at low velocity motion, the noise in the input data becomes significant. We use a linear model for predicting future mouse velocity to reduce noise interference.

The predicted mouse motion will then be mapped to the 3D virtual environment to determine the 3D motion of the viewer or any object in the environment. The initial results of such prediction method are very encouraging. Our experiments show that this method is more accurate than the popular prediction methods available.

We have implemented three methods for comparison with the actual user navigation movement:

actual - actual user navigation movement

P1 - our method

P2 - a predictor using the Guass-Markov process

P3 - a second order polynomial predictor

We tested the three methods by predicting 1, 3 and 5 steps ahead. The results are as follows:

Video of the actual navigation movement.

Video of the 1-step prediction results of the three methods.

Video of the 3-step prediction results of the three methods.

Video of the 5-step prediction results of the three methods.

In general, our motion prediction method is considered as a short-term prediction scheme. We are currently investigating a more long-term prediction scheme, which hopefully will further improve the accuracy of our prediction method.

On the other hand, we have noticed that finger motion has a similar motion characteristics as the mouse motion. We have recently extended the mouse motion prediction method for hand motion prediction by considering the static and dynamic constraints of finger motion.

We have performed two sets of experiments to evaluate the new hand motion prediction method. The first set of experiments compare the accuracy of the 4 predictors at different prediction lengths (how far ahead in time that we attempt to predict) as follows:

File predict_100ms.mov: 0.1s
File predict_300ms.mov: 0.3s
File predict_500ms.mov: 0.5s
The four predictors used in the above experiments are:

actual - actual user hand motion captured at the predicted time (reference)

SOP: the Second Order Polynomial predictor

GMM: the Gauss-Markov Model predictor

HM: the Hybrid Motion Model predictor (without constraints)

EHM: the Constraint-Based Hybrid Model predictor (the new method)

The second set of experiments show the application of the 4 predictors in a classic game called "Rock, Paper and Scissors". The following video demos compare the prediction accuracy of the 4 predictors during interactions. The hand on the left is a local hand (connected to the local host), while the hand on the right is a remote hand, which gesture is predicted. The prediction length was set at 0.5s.

File rps_actual.mov: using the original gestures (reference)
File rps_sop_500ms.mov: using the Second Order Polynomial predictor
File rps_gmm_500ms.mov: using the Gauss-Markov Model predictor
File rps_hm_500ms.mov: using the Hybrid Motion Model predictor
File rps_ehm_500ms.mov: using the Constraint-Based Hybrid Model predictor (the new method)

Recently, we have also started looking at the accuracy of motion prediction.

Publications:

Rynson Lau and Kenneth Lee, "On Error Bound Estimation for Motion Prediction," Proc. of IEEE VR, pp. 171-178, Mar. 2010.
Addison Chan, Rynson Lau, and Lewis Li, "Hand Motion Prediction in Distributed Virtual Environments," IEEE Trans. Visualization and Computer Graphics, 14(1):146-159, Jan. 2008.
Addison Chan, Rynson Lau, and Lewis Li, "Motion Prediction in Gesture-based Collaborative Design Environments," Proc. of IEEE VR, pp. 73-80, Mar. 2006.
Addison Chan, Rynson Lau, and Beatrice Ng, "Motion Prediction for Caching and Prefetching in Mouse-Driven DVE Navigation," To appear in ACM Transactions on Internet Technology.
Addison Chan, Rynson Lau, and Beatrice Ng, "A Hybrid Motion Prediction Method for Caching and Prefetching in Distributed Virtual Environments," Proceedings of ACM VRST, pp. 135-142, November 2001.
Addison Chan, Rynson Lau, and Antonio Si, "A Motion Prediction Method for Mouse-Based Navigation," Proceedings of Computer Graphics International (CGI), IEEE Computer Society Press, pp. 139-146, July 2001.

Motion Prediction for Distributed Virtual Environments

Last updated in March, 2010.