For the analysis, design, and operation of systems such as mobile communication systems, it is important to be able to predict the movement paths of humans. Most existing methods identify motion patterns (MPs) based on observed trajectories and then use them for prediction.
The authors present an approach based on classified MPs in terms of their credibility, which shows how indicative these patterns are for the specific environment. In the first introductory section, the authors specify the problem and present a comprehensive overview of existing approaches with adequate references. Section 2 provides a brief summary of the framework, which comprises four parts: trajectory extraction, MP clustering, MP classification, and motion prediction. In the extensive section 3, these parts are explained in detail.
The trajectories are extracted from video frames using a Bayesian framework and 3D human-shape models. Clustering uses the constrained gravitational method to generate MPs and general representations of a subgroup of trajectories. The credibility levels of the clustered MPs are based on the mass (number of trajectories) and size (distance between the bounding trajectories) information of each cluster. To predict a trajectory, an adaptive Bayesian probability model is proposed for the similarity measurement of this trajectory and the MP clusters. The credibility level of the cluster controls the number of time steps--that is, the higher the credibility, the more time steps are predicted. The whole procedure is well presented in detail, though some variables and terms could have been explained better.
Section 4 validates the proposed approach using experiments in both simulated and real-world scenes. Unfortunately, the simulation experiments are not clearly explained, which means that they are not really repeatable--a basic requirement for any kind of experiment. The paper ends with section 5’s conclusions. The experiments show that the proposed method is an improvement over the recursively applied autoregressive model.