Pedestrian trajectory prediction has a broad appeal in research due to its usefulness in many branches that go from aiding society like autonomous car driving or better city planning to saving human lives in case of disastrous events. The best approach that has been broadly accepted in the last years is through the use of Graph Neural Networks, thanks to their ability to represent the group of people with a selective granularity and the relation inside of it. This paper is going to explore this topic under the lenses of an office building floor during an evacuation. At first the final evacuation time will be predicted from a very coarse graph of the initial situation, then meshes will be introduce to raise the granularity of the prediction, and finally the trajectory of the whole event will be predicted. The model takes advantage of the SWIM algorithm, which offers an alternative to random weights and biases sampling that is also computationally efficient. The dataset from which to extrapolate the data is going to be simulated inside of the software Vadere, which will also be used to create the meshes. Predictions and results are evaluated, as well as a comparison through an animation of the model results versus the ground truth. Finally, ways to continue and expand what has been established in this project will be explored.     «

Pedestrian trajectory prediction has a broad appeal in research due to its usefulness in many branches that go from aiding society like autonomous car driving or better city planning to saving human lives in case of disastrous events. The best approach that has been broadly accepted in the last years is through the use of Graph Neural Networks, thanks to their ability to represent the group of people with a selective granularity and the relation inside of it. This paper is going to explore this...     »