Generating Versatile Training Samples for UAV Trajectory Prediction

Following the success of deep learning-based models in various sequence processing tasks, these models are increasingly utilized in object tracking applications for motion prediction as a replacement of traditional approaches. On the one hand, these models can capture complex object dynamics while requiring less modeling, but on the other hand, they depend on a large amount of training data for parameter tuning. Towards this end, an approach for generating synthetic trajectory data of unmanned-aerial-vehicles (UAVs) in image space is presented in this paper. Since UAVs are dynamical systems, they are bound to strict physical constraints and inputs for controlling. Thus, they cannot move along arbitrary trajectories. To generate executable trajectories, it is possible to apply solutions from trajectory planning for our desired purpose of generating realistic UAV trajectory data. Accordingly, with the prerequisite that UAV trajectories fulfill a smoothness criterion corresponding to a minimal change of higher-order motion, planning methods enabling aggressive quadrotor flights are applied to generate optimal trajectories through a sequence of 3D waypoints. By projecting these maneuver trajectories to image space, a versatile trajectory data set is realized. To demonstrate the applicability of the synthetic trajectory data, we show that deep learning-based prediction models solely trained on the synthetically generated data can outperform traditional reference models on a real-world UAV tracking dataset. The evaluation is done on the publicly available ANTI-UAV dataset.