The purpose of the Providentia++ project is to set up digital imagery and infrastructure along the German A9 highway in order to create a real time “digital twin” of the traffic. Ultimately, the goal is to also provide services to various stakeholders such as autonomous cars or smart traffic lights, among others. One obvious service is the prediction of short-term trajectories of traffic participants, which may be used for e.g. collision avoidance or lane change optimization in autonomous cars. In this report we present the implementation of a trajectory prediction framework within the Providentia++ backend. This framework allows researchers to easily export training data from the digital twin and to integrate models into the backend, while also providing real-time visualization and model performance metrics. As an implementation and performance reference, we integrated three models: two baseline models - a constant velocity model and a vanilla LSTM - and a state of the art motion prediction model, FloMo. The whole framework can be integrated into the live system using a docker image and be used to provide live visualization and metrics.
«
The purpose of the Providentia++ project is to set up digital imagery and infrastructure along the German A9 highway in order to create a real time “digital twin” of the traffic. Ultimately, the goal is to also provide services to various stakeholders such as autonomous cars or smart traffic lights, among others. One obvious service is the prediction of short-term trajectories of traffic participants, which may be used for e.g. collision avoidance or lane change optimization in autonomous cars....
»