Cross-View Geolocalization and Disaster Mapping with Street-View and VHR Satellite Imagery: A Case Study of Hurricane IAN

TL;DR

CVDisaster addresses the urgent need for rapid disaster response by jointly estimating geolocation and damage perception from cross-view imagery. It introduces a dual-model approach: CVDisaster-Geoloc uses a Siamese ConvNeXt encoder with contrastive learning for cross-view geolocalization, while CVDisaster-Est employs a CGCViT-based classifier to predict damage levels from paired Street-View and VHR satellite images. Validated on Hurricane IAN with the newly released CVIAN dataset, the framework achieves competitive performance and demonstrates the benefits of limited fine-tuning for rapid deployment. By releasing CVIAN and code, the work provides a practical, scalable path for GeoAI-enabled disaster mapping and situational awareness in GNSS-denied or degraded conditions.

Abstract

Nature disasters play a key role in shaping human-urban infrastructure interactions. Effective and efficient response to natural disasters is essential for building resilience and a sustainable urban environment. Two types of information are usually the most necessary and difficult to gather in disaster response. The first information is about disaster damage perception, which shows how badly people think that urban infrastructure has been damaged. The second information is geolocation awareness, which means how people whereabouts are made available. In this paper, we proposed a novel disaster mapping framework, namely CVDisaster, aiming at simultaneously addressing geolocalization and damage perception estimation using cross-view Street-View Imagery (SVI) and Very High-Resolution satellite imagery. CVDisaster consists of two cross-view models, where CVDisaster-Geoloc refers to a cross-view geolocalization model based on a contrastive learning objective with a Siamese ConvNeXt image encoder, and CVDisaster-Est is a cross-view classification model based on a Couple Global Context Vision Transformer (CGCViT). Taking Hurricane IAN as a case study, we evaluate the CVDisaster framework by creating a novel cross-view dataset (CVIAN) and conducting extensive experiments. As a result, we show that CVDisaster can achieve highly competitive performance (over 80% for geolocalization and 75% for damage perception estimation) with even limited fine-tuning efforts, which largely motivates future cross-view models and applications within a broader GeoAI research community. The data and code are publicly available at: https://github.com/tum-bgd/CVDisaster.

Figures (10)

• Figure 1: An overview of the proposed framework for Cross-view Geolocalization and Disaster mapping with street-view and satellite imagery, namely CVDisaster.
• Figure 2: The proposed framework, namely CVDisaster, addresses two key tasks simultaneously, which are 1) CVDisaster-Geoloc: cross-view disaster perception estimation using coupled Global Context Vision Transformer; 2) CVDisaster-Est: cross-view geolocalization via contrastive learning.
• Figure 3: The siamese image encoder for cross-view geolocalization using (a) a four-stage ConvNeXt; (b) the comparison of ConvNeXt and ReseNet blocks.
• Figure 4: Coupled Global Context Vision Transformer for Cross-View Imagery Classification and Damage Perception Estimation. The global token generator is highlighed.
• Figure 5: Overview of the study area together with the street-view and VHR satellite imagery. Five subareas are depicted in different colors in the middle image, where (a) to (d) are selected cross-view imagery pairs of CVDisaster.