Trajectory prediction by attention model considering pedestrians and vehicles interaction

Abstract

In order to realize safe automatic driving in urban areas where many kinds of moving objects move freely, it is necessary to predict the future behavior of moving objects. We propose a new trajectory prediction method based on the encoder-decoder framework, which can consider the interaction of moving objects in many aspects. By using labels and angular velocities for each moving object as inputs to the model and incorporating the interaction of moving objects using multi-head attention, the prediction accuracy is improved from the previous method.