Learning and Predicting Moving Object Trajectory: A Piecewise Trajectory Segment Approach

This paper presents an approach to predict future motion of a moving object based on its past movement. This approach is capable of learning object movement in an open environment, which is one of the limitions in some prior works. The proposed approach exploits the similarities of short-term movement behaviors by modeling a trajectory as concatenation of short segments. These short segments are assumed to be noisy realizations of latent segments. The transitions between the underlying latent segments are assumed to follow a Markov model. This predictive model was applied to two real-world applications and yielded favorable performance on both tasks.