The EUCENTRE Foundation

Eucentre is a private non-profit foundation that pursues a mission of research, training and service provision in the field of earthquake engineering and, more generally, of risk engineering.

Active in Pavia since 2003, it was established by four Founders, the University of Pavia, the University School for Advanced Studies IUSS of Pavia, the Italian Department of Civil Protection, the National Institute of Geophysics and Volcanology, to further develop the scientific, research and training expertise in the sector present in Pavia.

Today Eucentre has an important asset of experimental labs consisting of shaking tables able to reproduce any seismic event for testing both structural and non-structural elements and for the qualification of anti-seismic devices.

Eucentre operates within an international network with other research centres, earthquake engineering laboratories, institutions and companies.

Eucentre is Centre of Competence of the Italian Department of Civil Protection, to which it provides emergency support, elaboration of risk scenarios and research activities for the improvement of Civil Protection activities.

Today Eucentre is an international reference centre for institutions, for which it operates in the definition of emergency plans, in the elaboration of risk scenarios, in the vulnerability assessment of buildings and infrastructures, and for companies, to which it offers experimental and supporting services for the seismic design in various sectors.

EUCENTRE is supported by the Computer Vision & Multimedia Lab of University of Pavia. The CVM Lab is part of the Department of Electrical, Computer and Biomedical Engineering (DIII). Since 1976, CVMLab has gained substantial experience in automated image analysis in several fields like robotic vision, security surveillance, structural assessment, face recognition, medical imaging, advanced human-machine interfaces and digital humanities. More recently, CVMLab has specialized and focused on AI methods for computer vision and deep learning.

The Infrastructure Monitoring System

Within Teamaware project, EUCENTRE is responsible for the implementation if the Infrastructure Monitoring System, which is intended for the identification of risks and threats surrounding the first responders based on the visual detection of damages on structures and infrastructures in the critical event area, using drone surveillance.

The system can be used in any situation in which the structural integrity may be jeopardized. This means as a first use after earthquake or explosion, but also in case of landslides hitting infrastructures or fires on structures.

Two basic assumptions have to be kept in mind:

• The system is intended to be used at minimum by a drone operator and a payload operator. Both are suggested to be structural engineers, but the latter has to be.
• Result output from the tool is intended to be used by structural engineers in order to evaluate structural danger situations
• Onsite independent performance of the system shall be guaranteed

Specific procedures will be designed for structural damage detection and identification: based on visual data caught by drones, mostly acquired with optical sensors and in daytime, post-flight external inspection will be carried out on damaged structure with post-flight automated screening techniques. In addition, images/video collected by other means can be as well analysed. The main purpose of the automated screening process will be reducing the amount of data by several orders of magnitude to a reduced set of data that might be worth inspecting by human experts, thus enhancing the capability to assess and detect structure-critical situations.

The automated screening algorithm for damage recognition will be based on Deep Convolutional Neural Networks (DCNN) trained for item detection and localization.

The system will be designed for running offline (i.e. on non-embarked computing devices) and will be able to run on high-end portable computers (laptop) equipped with a suitable GPU.

Overall, the system will have three main components:

• Onsite inference video screening module, to be ran on a powerful laptop workstation.

• Offsite training and testing module, a composite software component that implements all required functions for the purpose in point.

• Image generation component, based mainly on Blender, for producing annotated images.

The main output produced by the system, intended as the output of the onsite inference module, will be a human-readable report describing the results of the screening process. More precisely, the report will contain sufficient information for human experts about where possible lesions have been detected in the input video, in the format of time intervals. A classification of the lesion type will also be provided, together with specific video frames with visual annotations.

Overall, the integration of the system with the TEAMAWARE Platform will be based on asynchronous communications. More precisely, the system will be designed to run onsite in an independent mode (i.e., drone plus laptop workstation) and to transfer the results of the automatic screening process on the platform at a later stage, as soon as a reliable connection can be established.