Presenter
Mr. Shun'an Zhou
Co-author(s)
Dr. Heng Lv, Prof. Chi Zhang
Organisation
Dalian University of Technology
Sub-theme
3. Building Resilience for Disaster Prevention and Mitigation
Topic
3-5. Monitoring and early warning of water-related disasters
Body
Detection of ponding levels timely and accurately during urban floods is the basis of effective disaster prevention and mitigation. New data sources such as social media and road surveillance videos record the process of urban floods, and the development of computer vision technology brings new opportunities for extracting ponding information from videos. This study proposes a computer vision-based method for automatically estimating the spatial-temporal distribution of ponding levels on urban streets from surveillance videos. Firstly, a dataset of images including sedans compiled from three sources was collected to train an object detection model, You Only Look Once vision 3. All the sedans occurring in the dataset were labeled with corresponding local ponding levels. The trained model was then used to identify the sedans from the videos and simultaneously estimate the ponding levels at the location of sedans, hence the sampling points with ponding level and spatial information were obtained. The number of sampling points can be flexibly changed through adjusting the video sampling interval. Secondly, a spatiotemporal outlier detection method was employed to detect and delete the outliers of ponding levels at each time step. Finally, the ponding level distribution from which we can obtain the real ponding distribution was further estimated through spatiotemporal interpolation method based on inverse distance weight from the updated sampling points. Our method was assessed in two pluvial flood events at a street crossing, Dongguan Street, in Dalian, China. The mean average precision of the trained model reached 78%, which verified the validity of the model. The ponding levels estimated by our method were validated with the submerged depth of a static reference. Furthermore, the outlier detection improved the accuracy of ponding level estimation to 88% on average in the cross-validation. The obtained ponding level distribution can be used to analyze the urban flood process and predict future flood trend, which contributes to upgrading urban drainage systems and improving flood prevention and mitigation in cities.