Synthetic Extreme Weather for AI Training: Concept and Validation

The development of machine learning based perception algorithms for safety-critical applications such as automated driving systems requires controlled environments for data collection as well as for testing under different scenarios and adverse weather conditions. In order to enable data collection and testing under controlled rain conditions, this paper proposes a methodology for generating three different synthetic rain intensities in an outdoor facility and discusses the validation of the results. The analysis is based on the comparison with real rain data and with indoor synthetic rain, utilizing data from a disdrometer, a weather station and square meter method. The emulated synthetic rain presents good results, with acceptable uniformity in the outdoor facilities and Rain Droplet Size Distribution similar to real rain data.