IMPLICITY: CITY MODELING FROM SATELLITE IMAGES WITH DEEP IMPLICIT OCCUPANCY FIELDS

For many safety-related applications such as road management, well-documented avalanches are important. To enlarge the information, webcams may be used. We propose supporting the mapping of avalanches from webcams with a machine learning model that interactively works together with the human. Relying on that model, there is a 90% saving of time compared to the "traditional" mapping. This gives a better base for safety-critical decisions and planning in avalanche-prone mountain regions.

Oftentimes when objective measurements are not possible, human estimates are used instead. In our study, we investigate the reproducibility of human judgement for size estimates, the mappings of avalanches from oblique photographs and remotely sensed imagery. The variability that we found in those estimates is worth considering as it may influence results and should be kept in mind for several applications.

Knowing where avalanches occur is very important information for several disciplines, for example avalanche warning, hazard zonation and risk management. Satellite imagery can provide such data systematically over large regions. In our work we propose a machine learning model to automate the time-consuming manual mapping. Additionally, we investigate expert agreement for manual avalanche mapping, showing that our network is equally as good as the experts in identifying avalanches.