Learning generative models for classification-recognition of human trajectories using semi-supervised EM algorithm

This work presents a semi-supervised EM algorithm for learning generative models for classification/recognition of human trajectories, with application to surveillance. The classifier is based on switched dynamical models, each model describing a specific motion regime. We present a semi-supervised modified version of the classical BaumWelch algorithm, which is able to take into account a subset of known model labels. The experimental results shows the effectiveness of the present approach in both synthetic and real data. It is shown as well, that the classifier learned with semi-supervision leads to a higher classification accuracy than the fully unsupervised version. This abstract describes the work presented in [2].