Automatic learning of 3D pose variability in walking performances for gait analysis

This paper proposes an action specific model which automatically learns the variability of 3D human postures observed in a set of training sequences. First, a Dynamic Programing synchronization algorithm is presented in order to establish a mapping between postures from different walking cycles, so the whole training set can be synchronized to a common time pattern. Then, the model is trained using the public CMU motion capture dataset for the walking action, and a mean walking performance is automatically learnt. Additionally statistics about the observed variability of the postures and motion direction are also computed at each time step. As a result, in this work we have extended a similar action model successfully used for tracking, by providing facilities for gait analysis and gait recognition applications.Peer Reviewe