Road Weather Condition Detection Based on Sub-Image Feature Fusion

Road weather conditions can serve as prior information to accurately estimate the tire-road friction coefficient. This information can also prevent autonomous vehicles from driving onto slippery roads, thereby enhancing driving safety. Unlike other object detection tasks, road weather condition detection requires identifying various states of the same object, where the features of these states are often very similar and challenging to distinguish. To address this issue, this article proposes a sub-image feature fusion (SiFF) module, which expands the receptive field of the feature map, allowing it to fully capture the environmental information surrounding the road. Under natural conditions, roads with different weather conditions exist in diverse environments. Therefore, acquiring more comprehensive environmental information can help improve the differentiation of similar features, thereby enhancing detection accuracy. To validate the proposed algorithm, we established a road weather condition dataset containing over 4000 images with high-quality pixel-level annotations. Experimental results show that incorporating the SiFF module into the existing segmentation algorithm can improve the mean IoU of road weather condition segmentation by approximately 3%, demonstrating the effectiveness of the proposed module.