Blog

The second HERON blogpost focuses on the Institute of Communications and Computer Systems (ICCS) and their role in the project.

Please describe your organization.

The Institute of Communications and Computer Systems (ICCS) is a non-profit academic research body established in 1989 by the Greek Ministry of Education to carry out research and development activities in the area of telecommunications, systems and techniques, computer systems and their applications in transceivers, radar, electromagnetic sensors, satellite and wireless communications, electromagnetic phenomena modelling, neural networks, systems, software and hardware engineering, telematics and multimedia applications, transport applications, control systems, biomedical engineering, and electric power.

What are some of the research capabilities of ICCS that can contribute to the work in HERON?

The Laboratory of Photogrammetry, Signal Processing & Computer Vision performs research in spectral analysis, signal processing, machine learning, and environmental applications. It is equipped with GIS-based tools (ArcGis, QGis), along with the respective software plug-ins (e.g., GeoHealth). Moreover, the laboratory is equipped with capturing devices, such as ultra-high resolution hyperspectral cameras, thermal cameras, infrared cameras, multi-spectral cameras, many optical RGB cameras for surveillance purposes, 3D laser scanners, structured light scanners, multi-view camera architectures, Time of Flight sensors for omnidirectional analysis, photographic machines of high resolutions, high-speed cameras. Lastly, it also has all the software for processing hyper-spectral data and 3D data and generating 3D models.

What is the role of ICCS in HERON?

ICCS is the HERON Project Coordinator and leader of WP3, which is related to various AI-based algorithms and tools for recognition, classification, and localization of the HERON points of interest. ICCS is also Responsible for the AI and deep machine learning algorithms, as well as the computer vision tools that will be integrated into the HERON robotic platform.

What are some of the results you have already achieved in HERON?

Various state-of-the-art computer vision tools (e.g., YOLO algorithms for object detection problems and U-Net models with recurrent residual and attention modules as well as deep transformer networks for precise segmentation tasks) have been analyzed, implemented, and compared for the effective classification, localization, and segmentation of the various HERON points of interest.

For instance, numerous deep learning tools have been designed for crack segmentation, pothole localization, blurred road marking detection, traffic cone identification, traffic sign classification, and RUP (Removable Urban Pavement) recognition. Various road infrastructure-related datasets that derive from sensors mounted on inspection vehicles or drones have been acquired and annotated, and are now accessible to the scientific community, for verifying the results and further research. Regarding these applications indicative experimental results have been presented in the following scientific publications:

• https://dl.acm.org/doi/abs/10.1145/3529190.3534746
• https://link.springer.com/chapter/10.1007/978-3-031-17601-2_37
• https://link.springer.com/chapter/10.1007/978-3-031-17601-2_34
• https://dl.acm.org/doi/abs/10.1145/3594806.3596560
• https://link.springer.com/chapter/10.1007/978-3-031-47969-4_16

What is the exploitation potential of these results?

Of all public assets, road infrastructure tops the list. Roads are crucial for economic development and growth, providing access to education, health, and employment. The maintenance, repair, and upgrade of roads are therefore vital to road users’ health and safety as well as to a well-functioning and prosperous modern economy. Recent studies indicate that artificial intelligence plays a crucial role in the automated visual inspection and maintenance of road infrastructures. In pursuit of this objective, the potential for exploitation can be associated with various key aspects, including:

• Algorithms that are responsible for transforming the 2D captured visual cues of the sensors into high-level feature maps capable of modelling and representing the various points of interest using AI-driven and deep learning methodologies.
• Machine learning methods that combine textural, geometric and semantic information in a high-level fused manner, to recognize, classify and localize the points of interest.
• Geo-reference and precisely 3D localize the points of interest.

What are some of the activities/projects related to HERON, in which your organization is involved?

The Laboratory of Photogrammetry, Signal Processing & Computer Vision has been involved in the following activities/projects that are linked to HERON:

• Defect detection in civil infrastructures (H2020 PANOPTIS project),
• Development of crack detection algorithms using computer vision tools for tunnels (FP7 Robo-Spect project),
• Artificial intelligence algorithms for processing thermal data for security purposes in critical civil infrastructures (H2020 STOP-IT project),
• Detection and 3D modeling of civil infrastructures after natural disasters such as earthquakes (FP7 INACHUS project),
• 3D modeling and reconstruction of defects in prominent cultural heritage monuments, buildings, and historic areas affected by climate change (H2020 HYPERION project),
• Pipeline inspections for defects using computer vision tools (FP7 ZoneSec Project),
• Evacuation methods and algorithms using computer vision and machine learning for large-scale critical environments (FP7-eVacuate project),
• Network of fellows investigating the impact of climate change on cultural heritage structures (H2020 YADES project),
• Real-time signal adaptive/tracing algorithms from un-calibrated capturing sensors equipped with technologies extracting geolocations of detected objects of interest (FP7-SCOVIS project).