Developing an sensor agnostic Infrastructure-Based Control Approach for Crossroad Assistance using Simulation

Crossroad assistance has the potential to contribute to a fair distribution of road space and increase the safety of all road users. This paper presents an infrastructure-based control approach to support crossroad assistance by implementing remote-controlled turning maneuvers on a test track. The approach involves several steps, including the detection of traffic events using camera-based object recognition, inserting dynamic objects into the simulation via V2X transmission, adding lanes and traffic signal phases through V2X data forwarding to a high-precision simulation map, trajectory prediction of all dynamic objects in the simulation, sending a trajectory list to the test vehicle, and providing feedback through the vehicle's response to the trajectory list and re-detection of objects. The study faced some difficulties in reliably tracking objects over several time steps due to issues such as occlusion and blind spots at the intersection. Simple approaches, such as applying a Kalman filter to extrapolate an object's motion vector, are not sufficient in more complex scenarios such as crossroads. Traditional methods may fail to identify an object with its previous ID after occlusion if the object has changed direction during the occlusion, which frequently occurs during turning maneuvers. Tracking supported by SUMO can address this problem because the simulation knows the mentioned directional changes and driving lanes. The use of CPM for transmitting object lists makes the simulation independent of the sensor. Thus, it is possible to replace or extend the object detection system with another solution. This infrastructure-based approach can support crossroad assistance by improving traffic flow and increasing the safety of all road users