Pedestrian motion classification on omnidirectional treadmill

In this paper, the direction integrated pedestrian motion classification method on the omnidirectional treadmill is proposed based on a navigation algorithm. The virtual reality technology is widely applied to a military training in recent years since previous drill conducted outside is relatively cost and time inefficient. Among several roles in training system, motion recognition including direction determination is essential, but the classification result by a classifier only becomes a problem. In order to improve the classification accuracy, navigational error is obtained using an EKF-ZUPT algorithm, and the direction is estimated from the corrected position by previous states. Aside from the determined direction, features are extracted, and the learning and collection steps are conducted. The final recognition results are acquired from the combination of direction and a classifier. The experimental results show that motion classification accuracy of the proposed algorithm has over 90%.