For mobile robotic systems to navigate smoothly and cooperate with pedestrians in dynamic environments, the ability to predict where pedestrians will move in the next few seconds is crucial. To tackle this problem, many solutions have been developed which take the environment’s influence on human navigation behavior into account. Few however, do this in a way which seamlessly generalizes to new environments where no prior observation of pedestrians is available. In this thesis, we propose a novel method that uses convolutional neural networks (CNNs) to predict local statistics about the direction of likely human motion. Specifically the network takes crops of a discretized floor plan as an input and computes a cell-wise transition probabilities over exit directions conditioned on the entry direction into the cell. These environment-aware motion models can be used to improve the tracking and prediction performance of existing multi-target tracking methods. We incorporated the learned motion models into two different existing multi-target tracking methods, Bayesian Occupancy Filter (BOF) and Particle Based Bayesian Occupancy Filter (PBOG). We evaluate the prediction performance of these tracking methods on real-world pedestrian data and demonstrate that the CNN predicted motion model considerably improves prediction quality of the tracking methods in previously unseen validation environments. Even though we propose to train the CNN entirely in simulation, our experiments suggest that the learned models generalize to real-world pedestrian data that we used in our evaluation.