Adverse Weather Benchmarks and Dataset for Object Detection in Autonomous Driving

Advancements in flexible mobility solutions are directing efforts towards achieving autonomous electric driving in transport and logistics domains. Autonomous systems offer increased availability, flexibility and decreasing costs. To ensure safe and reliable behavior in autonomous driving, it is imperative to employ object detection techniques based on sensor information. Despite the significance of these techniques, limited research has been conducted to compare and assess their efficacy in adverse weather conditions like rain, fog, or snow. This paper addresses this gap by evaluating segmentation techniques on a new benchmark dataset that includes various weather scenarios. The findings underscore the impact of weather conditions on the performance of the evaluated methods, emphasizing the importance of the proposed benchmark and the analysis of solutions regarding adverse weather conditions to ensure safe and reliable behavior of autonomous vehicles.