Real-time rain severity detection for autonomous driving applications

Rain produces a complex set of visual effects on an image captured by a camera sensor. Hence, the performance of any vision-based system gets adversely affected due to rain. Severity of impact is generally proportional to severity of rain. Such an effect is particularly relevant for advanced driver assistance systems (ADAS) and autonomous driving (AD) scenarios given the criticality of environment perception in such scenarios. In order to mitigate the adverse impact of such interference, detection and severity classification of rainy conditions is an essential part of the sensor support system. In this context, we present a novel rain detection and severity classification algorithm based on a neural network with inputs such as localized DCT and image-based features. The proposed approach is tested on real data set collected on the road and subjectively quantified with respect to rain severity. The method performs well in terms of both detection and classification accuracy. The proposed algorithm also achieves real-time operational capability owing to its low complexity.