The field of driver assistance systems faces new technologies like vehicle-to-x (V2X) communication [1] and cloud services to improve safety on the streets. These upcoming functions carry new possibilities and new challenges as well. Thanks to scaling-up techniques, it is already possible to gather and manage huge amounts of data in the cloud with less time consumption compared to standard systems. This data is suitable to be used for statistics and pattern recognition based functions with time-critical demands.
But beside the development of these functions there is always a need for evaluation to assure correct functionality. Therefore, IAV created a concept to dynamically evaluate cloud-based active safety systems like the wrong-way driver detection function developed by IAV, which is conceived to warn oncoming vehicles on motorways. This goal is achieved with cloud and V2X techniques.
The wrong-way driver warning is a service running on the cloud, which receives V2X messages from vehicles including their positioning data, heading and other information to reference their traces via map matching [2] algorithms. An underlying database compares the data with continuously calculated thresholds to recognize wrong-way drivers and warn others in the surrounding areas via V2X against them. As changing infrastructures like construction sites have an effect on the false-positive rate of the classification, it needs to be dynamically tested if the function is able to react appropriately.
For this reason we propose an approach for an evaluation process based on an integrated self-controlling module. This module notices changes in trajectories which can be caused by changed traffic guidance and adapts the detection parameters. As a result, the following detection should classify the vehicles correctly based on the changed conditions.
The discussed process of the introduced wrong-way driver warning shows an example how a specific cloud- based function can be dynamically evaluated without dependence on the functional logic. Especially functions whose results strongly depend on ever-changing input data need solutions to test their behavior in critical situations. This approach delivers also a possible answer how to implement additional methods to create more flexible and long-term reliable cloud-based active safety systems.