Trajectory prediction in the realm of sparse data

• Yasser Taha

Predicting the future movement of objects in video sequences is still an active area of research. In this work, we focus on predicting the future movements of pedestrian trajectories in the realm of sparse data. Four cornerstone novel approach algorithms that rely on Markov Models were developed to overcome sparsity and capture the complex behaviour of pedestrians. The algorithms presented infers the future path in O(Ntrq) time complexity, where Ntrq is the number of points in the trajectory query. The prediction accuracy will be shown to reach up to 95% accuracy on the test set. A detailed comparison between this approach and state-of-the-art approaches overcomes some of the addressed limitations in the literature. Furthermore, a novel approach will be illustrated that is able to indirectly infer the topology of the road by inferring the most likely path between the exits in a scene.