Evaluation of filtering methods for the prediction of human position during walking by means of kinematical models

In a setting shared by humans and machines, the short-term prediction in real time of a person's position during a walking displacement becomes necessary from a real and perceived safety point of view. Classical kinematical models inspired by the Newton's laws based on the actual position and orientation of the subject have been proposed to estimate her future position. However, we have found that the natural alternation of the step during human gait forces fluctuations in the position and orientation of the walking subject that limit the predictive capabilities of the kinematical models. We check whether filtering approaches may be valid to remove the fluctuations in real time. As an alternative, we propose to exploit biomechanical factors of human gait, in particular to combine the sampled position and orientation of the subject with her hip orientation to correct these fluctuations thus permitting the models to be applied in a real-time prediction context.