Deep Sensing of Urban Waterlogging

Shi-Wei Lo,Jyh-Horng Wu,Jo-Yu Chang,Chien-Hao Tseng,Meng-Wei Lin,Fang-Pang Lin
DOI: https://doi.org/10.48550/arXiv.2103.05927
2021-08-16
Abstract:In the monsoon season, sudden flood events occur frequently in urban areas, which hamper the social and economic activities and may threaten the infrastructure and lives. The use of an efficient large-scale waterlogging sensing and information system can provide valuable real-time disaster information to facilitate disaster management and enhance awareness of the general public to alleviate losses during and after flood disasters. Therefore, in this study, a visual sensing approach driven by deep neural networks and information and communication technology was developed to provide an end-to-end mechanism to realize waterlogging sensing and event-location mapping. The use of a deep sensing system in the monsoon season in Taiwan was demonstrated, and waterlogging events were predicted on the island-wide scale. The system could sense approximately 2379 vision sources through an internet of video things framework and transmit the event-location information in 5 min. The proposed approach can sense waterlogging events at a national scale and provide an efficient and highly scalable alternative to conventional waterlogging sensing methods.
Computer Vision and Pattern Recognition,Artificial Intelligence,Computers and Society
What problem does this paper attempt to address?
The paper attempts to address the issue of how to monitor and locate flooding events in urban flood disasters in real-time through an efficient large-scale visual perception system, in order to provide timely disaster information, thereby mitigating the impact of floods on socio-economic activities and enhancing public safety awareness. Specifically, the paper proposes a visual perception method based on deep neural networks and information communication technology, aiming to achieve the following goals: 1. **Real-time monitoring**: During the typhoon season, quickly and accurately monitor the flooding situation across the entire island. 2. **Event localization**: Collect and process a large number of video data sources through the Internet of Video Things (IoVT) framework to achieve real-time localization of flooding events. 3. **Efficient transmission**: Transmit event location information to users within 5 minutes, providing efficient disaster management support. 4. **Large-scale application**: The system can perceive flooding events on a national scale, offering an efficient and scalable alternative to traditional flood monitoring methods. By integrating deep learning technology and information communication technology, the paper develops an end-to-end visual perception system capable of processing large amounts of video data in a short time, identifying flooding events, and providing real-time information services. This system not only improves the efficiency of disaster response but also enhances public awareness and response capabilities to flood disasters.