Monocular Depth Estimation Algorithm for Rainy Scenes
Jinglun Hu, Zhuang Cao · 2024
In rainy environments, rainwater significantly reduces the visibility of scene objects, which in turn affects the accuracy of monocular depth estimation tasks. Moreover, under rainy conditions, obtaining large-scale scene depth ground truth is very difficult. To address these issues, this paper proposes a self-supervised joint learning network for monocular depth estimation in rainy scenes, which can achieve high accuracy without the need for depth ground truth labels. The network integrates a depth estimation sub-network and a rain removal sub-network. Furthermore, this paper also establishes a rain imaging model that comprehensively considers the effects of rain lines and rain fog. Based on this model and the outdoor photo dataset DrivingStereo, this paper also creates a rainy image dataset named RainDrivingStereo for training and verifying the network of this paper. Through experiments on the RainDrivingStereo dataset as well as other synthetic and real datasets, this paper fully proves the superiority of the proposed network.