Deep-Learning Rain Detection and Wiper Control Featuring Gaussian Extraction

The automatic wiper system detects the pattern and the amount of water on the windshield and then determines to activate the wipers to maintain visibility through the front windshield. In the past, the visual rain detection system based on the deep convolutional neural network was trained from the holistic front views captured by the dashcam. The method closely reflects the driver's visual experience and outperforms the existing state-of-the-art. In this article, we further propose a technique to enhance the training data and improve detection performance. Specifically, the composition of the training image set is improved by Gaussian position extraction to focus more on the driver's front view and suppress the contents that are irrelevant near the sides. Experimental evidence shows that the proposed data augmentation method enhances the rain detection recall rates by more than 10%. Such improvement was confirmed by both of the existing dataset and a new one we created for testing.