Detecting Safe Routes During Floods Using Deep Learning

Background

Semi-automatic road extraction from digital images by Hamid Reza Riahi Bakhtiari and Abolfazl Abdollahi (2017) uses a combination of edge detection and Support vector machines to extract different types of roads. The proposed system uses canny edge detection followed by a Support Vector Machine (SVM) classifier to extract roads. The approach proposed by them is semi-automatic which means they require human intervention for the classifier to work, which is not ideal for the scenario proposed in this paper. To address this, a U-net Convolutional model is proposed in this paper to extract the roads from the satellite image without any human intervention. This extracted roadmap is then compared with existing maps to give the summary of roads available.

Automatic extractions of road intersections from satellite imagery in urban areas by Ali, Boshir, A., & Ariful (2010) proposes a road extraction algorithm using morphological direction filtering to extract the road layers from other layers and extract the road intersections to determine road orientation and interconnectivity. This method promises a high accuracy in road extraction.

Krishna Kant Singh and Akansha Singh (2017) propose a system to identify flooded areas from satellite images using the Hybrid Kohonen Fuzzy C-Means sigma. The proposed system is accurately able to detect flooded areas by comparing the images of flooded areas with ground truth images. The system is accurate and efficient but is unable to provide insight into the situation of intensity and the damage the flood has caused to roads.

Flood evacuation and rescue: The identification of critical road segments using whole-landscape features by Edward Helderop and Tony H. Grubesic (2019) proposes a method for detecting critical road segments by a moving window/grid approach in a post-disaster landscape by treating the road as a network. The above approach gives a good idea of what road segments are important but is unable to predict the road condition during floods. If a critical road segment gets submerged in the water there is no way of knowing what other pathways can be used to reach to and from the flooded area.