RNN-based Prediction of Pedestrian Turning Maneuvers
The dynamics of objects, such as pedestrians, varies over time. Commonly this problem is tackled with traditional approaches like the Interacting Multiple Model (IMM) filter using a Bayesian formulation. Following the current trend towards using deep neural networks, in this paper an RNN-based alternative solution for pedestrian maneuver prediction is presented. Similar to an IMM filter solution, the presented model assigns a confidence value to a performed dynamic and, based on them, puts out a multi-modal distribution over future pedestrian trajectories. The qualitative evaluation is done on synthetic data, reflecting prototypical pedestrian maneuvers.