Deep learning algorithms are becoming the state-of-the-art technology used in the perception functions of AD and ADAS. With the safety assurance of level 5 AD systems the perception of the environment and especially dynamic objects becomes a central component. Data-driven AI algorithms are to a large extent determined by the training dataset and need to be tested and validated especially for safety critical edge cases which can be derived from real world scenarios that lead to accidents caused by human drivers. However, current available public data is limited to normal driving situations.
The PREPER dataset that is now publicly available strives to fill this gap by providing data of 160 pre-crash sequences involving two passenger cars in rural and city scenarios recorded at the AstaZero track in December 2020 and January 2021. It comes with a set of collision warning indicators for evaluating the object detection performance of AI perception systems with respect to safety requirements. The goal of this unique dataset is to support researchers to have easy access and use of validation data while triggering research and development on safety aspects of AI perception systems and eventually help improve the robustness of AI perception systems used in AD/ADAS. The project is supported by the “Open Research Programme” by Chalmers, RISE, and SAFER in collaboration with AstaZero and by the project “Safety-driven data labelling platform to enable safe and responsible AI” funded by the FFI Road Safety and Automated Vehicles programme. The dataset can be downloaded from https://github.com/AsymptoticAI/PREPER. For details please contact email@example.com.
A second dataset containing public road data is planned to be published subsequently complementing the dataset released today. For further information please send an e-mail to firstname.lastname@example.org or email@example.com.