With the intensification of global climate change, extreme precipitation events are occurring more frequently, making the monitoring and management of urban flooding a critical global issue. Urban surveillance camera sensor networks, characterized by their large-scale deployment, rapid data transmission, and low cost, have emerged as a key complement to traditional remote sensing techniques. These networks offer new opportunities for high-spatiotemporal-resolution urban flood monitoring, enabling real-time, localized observations that satellite and aerial systems may not capture. However, in low-light environments—such as during nighttime or heavy rainfall—the image features of flooded areas become more complex and variable, posing significant challenges for accurate flood detection and timely warnings. To address these challenges, this study develops an imaging model tailored to flooded areas under low-light conditions and proposes an invariant feature extraction model for flooding areas within surveillance videos. By using extracted image features (i.e., brightness and invariant features of flooded areas) as inputs, a deep learning-based flood segmentation model is built on the U-Net architecture. A new low-light surveillance flood image dataset, named UWs, is constructed for training and testing the model. The experimental results demonstrate the efficacy of the proposed method, achieving an mRecall of 0.88, an mF1_score of 0.91, and an mIoU score of 0.85. These results significantly outperform the comparison algorithms, including LRASPP, DeepLabv3+ with MobileNet and ResNet backbones, and the classic DeepLabv3+, with improvements of 4.9%, 3.0%, and 4.4% in mRecall, mF1_score, and mIoU, respectively, compared to Res-UNet. Additionally, the method maintains its strong performance in real-world tests, and it is also effective for daytime flood monitoring, showcasing its robustness for all-weather applications. The findings of this study provide solid support for the development of an all-weather urban surveillance camera flood monitoring network, with significant practical value for enhancing urban emergency management and disaster reduction efforts.

Winter road snow significantly alters surface friction conditions and traffic capacity, serving as a critical factor contributing to traffic accidents, congestion, and temporary traffic control measures. Compared with sparsely deployed road sensors and labor-intensive field inspections, traffic surveillance cameras offer advantages such as dense spatial coverage, low deployment cost, and continuous observation capability, providing a feasible solution for segment-level winter road condition monitoring. To meet traffic management needs, this study categorizes the impact of road snow on passability into four classes: Clear, Light, Medium, and Heavy. A road snow coverage dataset containing 10,498 images under complex traffic scenarios was constructed and has been publicly released. Furthermore, nine representative deep learning models were systematically evaluated to compare their recognition performance and applicability for this task. Experimental results show that all models achieved over 89% classification accuracy on the test set. To further examine cross-regional generalization capability, 48 surveillance cameras from Canada and Norway were selected for real-world validation. Among all models, Swin Transformer achieved the highest accuracy of 81.2% under complex lighting conditions and varying viewpoints, demonstrating superior stability and transferability. The findings provide quantitative guidance for model selection and engineering deployment of camera-based winter road monitoring systems.