The activities of WP4 are focused on the development of the Infrastructure Monitoring System (IMS), with
the objective to identify risks and threats surrounding the first responders based on the visual detection of
damages on structures and infrastructures in the critical event area, using advanced Machine Learning
algorithms for anomalies detection.
Thanks to drone surveillance of buildings and infrastructures, IMS will support expert engineers in the
identification of structural damages on video footage to contribute to the emergency management system,
providing information on the typology of damages detected and annotation of their location on significant
video frames in order to identify potential situations of risks during emergency operations.
The most common scenarios faced by the first responders after natural catastrophes, such as earthquakes
and landslides, have been studied to carried out a proper selection of the damage typologies to be
detected by the IMS. Moreover, the WP4 activities have been focused on the first design of the IMS
architecture, to provide insight about the technical specification of the system components, of the
functionalities to implement and the integration of the IMS with the global TeamAware architecture and
infrastructure. The user needs and the overall platform requirements have been analysed in order to meet
current and future needs considering the inputs of the experts of the consortium.
Regarding the deep learning algorithms to be implemented in the system, the most recent advances in the
state of the art have been analysed to select the most suitable software architecture for automated
damage detection. Furthermore, because the algorithms require a significant amount of data for training
and validation, the availability of suitable open-source datasets has been explored, in conjunction with the
development of a methodology for the generation of a dataset of semi-synthetic images for the purposes
of WP4. The latter dataset is created starting from 3D point cloud and mesh models of real buildings and
infrastructures obtained with photogrammetric techniques from drones surveys and it will be used for
algorithm training and validation together with the EUCENTRE’s dataset of annotated damages already
available.
