An embedded 3D human motion capture using the prediction provided from a walking model

We propose a human motion capture system embedded on a walker. It is constituted by a SwissRanger 3D camera and two infrared distance sensors. This paper deals especially with an original prediction module which is added to the observation system in order to improve its accuracy. A Kalman filter is used to estimate some parameters of the gait, the step length and step period. They are used to predict the next positions of the feet and a reference trajectory for the Zero Moment Point (ZMP). We deduce a future generalized position of the model of walking by taking into account these data. The trajectories of joints are then reconstructed by using an anatomical model from the Humanoid Motion Analysis and Simulation (HuMAnS) toolbox. In order to validate our approach, the result obtained with our embedded system is compared with ones obtained using Codamotion system. The experimental results shows that, as expected, the prediction module using a walking model in the monitoring scheme reduces the error values.